Search

Image compression is desired for many image-related applications especially for network-based applications with bandwidth and storage constraints. The face recognition community typical reports concentrate on the maximal compression rate that would not decrease the recognition accuracy. In general, the wavelet-based face recognition methods such as EBGM (elastic bunch graph matching) and FPB (face pattern byte) are of high performance but run slowly due to their high computation demands. The PCA (Principal Component Analysis) and LDA (Linear Discriminant Analysis) algorithms run fast but perform poorly in face recognition. In this paper, we propose a novel face recognition method based on standard image compression algorithm, which is termed as compression-based (CPB) face recognition . First, all gallery images are compressed by the selected compression algorithm. Second, a mixed image is formed with the probe and gallery images and then compressed. Third, a composite compression ratio (CCR) is computed with three compression ratios calculated from: probe, gallery and mixed images. Finally, the CCR values are compared and the largest CCR corresponds to the matched face. The time cost of each face matching is about the time of compressing the mixed face image. We tested the proposed CPB method on the “ASUMSS face database” (visible and thermal images) from 105 subjects. The face recognition accuracy with visible images is 94.76% when using JPEG compression. On the same face dataset, the accuracy of FPB algorithm was reported as 91.43%. The JPEG-compressionbased (JPEG-CPB) face recognition is standard and fast, which may be integrated into a real-time imaging device.

The current computer-aided detection (CAD) methods are not sufficiently accurate in detecting masses, especially in dense breasts and/or small masses (typically at their early stages). A small mass may not be perceived when it is small and/or homogeneous with surrounding tissues. Possible reasons for the limited performance of existing CAD methods are lack of multiscale analysis and unification of variant masses. The speed of CAD analysis is important for field applications. We propose a new CAD model for mass detection, which extracts simple Haar-like features for fast detection, uses AdaBoost approach for feature selection and classifier training, applies cascading classifiers for reduction of false positives, and utilizes multiscale detection for variant sizes of masses. In addition to Haar features, local binary pattern (LBP) and histograms of oriented gradient (HOG) are extracted and applied to mass detection. The performance of a CAD system can be measured with true positive rate (TPR) and false positives per image (FPI). We are collecting our own digital mammograms for the proposed research. The proposed CAD model will be initially demonstrated with mass detection including architecture distortion.

Unauthorized vehicles become an increasing threat to US facilities and locations especially overseas. Vehicle detection is a well-studied area. However, vehicle identification and intension analysis have not been sufficiently investigated. We propose to use multispectral (visible, thermal) images (1) to match the vehicle types with the registered (or authorized) vehicle types; (2) to analyze the vehicle moving patterns, (3) and study methods to utilize open information such as GPS and traffic information. When a vehicle is either permitted to access to the facility, or subjected to further manual inspection (scrutiny), the additional information (e.g., text) can be compared against the imagery features. We use information fusion (at image, feature, and score level) and neural network to increase vehicle matching accuracy. For the vehicle moving patterns, we will classify them as “normal” and “abnormal” by using driving speed, acceleration, stop, zig-zag, etc. The methods would support directions in physical and human-based sensor fusion, patterns of life (POL) analysis, and contextual-enhanced information fusion.

Image compression is necessary for data transportation, which saves both transferring time and storage space. In this paper, we focus on our discussion on lossy compression. There are many standard image formats and corresponding compression algorithms, for examples, JPEG (DCT — discrete cosine transform), JPEG 2000 (DWT — discrete wavelet transform), BPG (better portable graphics) and TIFF (LZW — Lempel-Ziv-Welch). The image quality (IQ) of decompressed image will be measured by numerical metrics such as root mean square error (RMSE), peak signal-to-noise ratio (PSNR), and structural Similarity (SSIM) Index. Given an image and a specified IQ, we will investigate how to select a compression method and its parameters to achieve an expected compression. Our scenario consists of 3 steps. The first step is to compress a set of interested images by varying parameters and compute their IQs for each compression method. The second step is to create several regression models per compression method after analyzing the IQ-measurement versus compression-parameter from a number of compressed images. The third step is to compress the given image with the specified IQ using the selected compression method (JPEG, JPEG2000, BPG, or TIFF) according to the regressed models. The IQ may be specified by a compression ratio (e.g., 100), then we will select the compression method of the highest IQ (SSIM, or PSNR). Or the IQ may be specified by a IQ metric (e.g., SSIM = 0.8, or PSNR = 50), then we will select the compression method of the highest compression ratio. Our experiments tested on thermal (long-wave infrared) images (in gray scales) showed very promising results.

It is a common knowledge that we see th e world because our eyes can perceive an optical image. A digital camera seems a good example of simulating the eye imaging system. However, the signal sensing and imaging on human retina is very complicated. There are at least five layers (of neurons) along the signal pathway: photoreceptors (cones and rods), bipolar, horizontal, amacrine and ganglion cells. To sense an optical image, it seems that photoreceptors (as sensors) plus ganglion cells (converting to electrical signals for transmission) are good enough. Image sensing does not require ununiformed distribution of photoreceptors like fovea. There are some challenging questions, for example, why don’t we feel the “blind spots” (never fibers exiting the eyes)? Similar situation happens to glaucoma patients who do not feel their vision loss until 50% or more nerves died. Now our hypothesis is that human retina initially senses optical (i.e., Fourier) spectrum rather than optical im age. Due to the symmetric property of Fourier spectrum the signal loss from a blind spot or the dead nerves (for glaucoma patients) can be recovered. Eye logarithmic response to input light intensity much likes displaying Fourier magnitude. The optics and structures of human eyes satisfy the needs of optical Fourier spectrum sampling. It is unsure that where and how inverse Fourier transform is performed in human vision system to obtain an optical image. Phase retrieval technique in compressive sensing domain enables image reconstruction even without phase inputs. The spectrum-based imaging system can potentially tolerate up to 50% of bad sensors (pixels), adapt to large dynamic range (with logarithmic response), etc. Keywords : Human vision, retinal imaging, retinal spectrum sensing, Fourier transform, image reconstruction.

The performance of face recognition can be improved using information fusion of multimodal images and/or multiple algorithms. When multimodal face images are available, cross-modal recognition is meaningful for security and surveillance applications. For example, a probe face is a thermal image (especially at nighttime), while only visible face images are available in the gallery database. Matching a thermal probe face onto the visible gallery faces requires crossmodal matching approaches. A few such studies were implemented in facial feature space with medium recognition performance. In this paper, we propose a cross-modal recognition approach, where multimodal faces are cross-matched in feature space and the recognition performance is enhanced with stereo fusion at image, feature and/or score level. In the proposed scenario, there are two cameras for stereo imaging, two face imagers (visible and thermal images) in each camera, and three recognition algorithms (circular Gaussian filter, face pattern byte, linear discriminant analysis). A score vector is formed with three cross-matched face scores from the aforementioned three algorithms. A classifier (e.g., k-nearest neighbor, support vector machine, binomial logical regression [BLR]) is trained then tested with the score vectors by using 10-fold cross validations. The proposed approach was validated with a multispectral stereo face dataset from 105 subjects. Our experiments show very promising results: ACR (accuracy rate) = 97.84%, FAR (false accept rate) = 0.84% when cross-matching the fused thermal faces onto the fused visible faces by using three face scores and the BLR classifier.

Humans can recognize a face with binocular vision, while computers typically use a single face image. It is known that the performance of face recognition (by a computer) can be improved using the score fusion of multimodal images and multiple algorithms. A question is: Can we apply stereo vision to a face recognition system? We know that human binocular vision has many advantages such as stereopsis (3D vision), binocular summation, and singleness of vision including fusion of binocular images (cyclopean image). For face recognition, a 3D face or 3D facial features are typically computed from a pair of stereo images. In human visual processes, the binocular summation and singleness of vision are similar as image fusion processes. In this paper, we propose an advanced face recognition system with stereo imaging capability, which is comprised of two 2-in-1 multispectral (visible and thermal) cameras and three recognition algorithms (circular Gaussian filter, face pattern byte, and linear discriminant analysis [LDA]). Specifically, we present and compare stereo fusion at three levels (images, features, and scores) by using stereo images (from left camera and right camera). Image fusion is achieved with three methods (Laplacian pyramid, wavelet transform, average); feature fusion is done with three logical operations (AND, OR, XOR); and score fusion is implemented with four classifiers (LDA, k-nearest neighbor, support vector machine, binomial logical regression). The system performance is measured by probability of correct classification (PCC) rate (reported as accuracy rate in this paper) and false accept rate (FAR). The proposed approaches were validated with a multispectral stereo face dataset from 105 subjects. Experimental results show that any type of stereo fusion can improve the PCC, meanwhile reduce the FAR. It seems that stereo image/feature fusion is superior to stereo score fusion in terms of recognition performance. Further score fusion after image/feature fusion does not benefit much (or at all) for performance improvement which may imply that the power of image/feature/score fusion shall not be doubly applied to a face recognition process pipeline.

A face recognition system usually consists of one recognition algorithm by processing single spectral images. For example, face pattern byte (FPB) algorithm was initially created using thermal (LWIR) images, while Elastic Bunch Graphic Matching (EBGM) algorithm was originated with visible (RGB) images. When there are two or more recognition algorithms and/or spectral images available, system performance can be enhanced using information fusion. In this paper, a score fusion with multispectral images is proposed to improve system performance, which is termed as an integrated multispectral face recognition system. Score fusion actually combines several scores from multiple matchers (algorithms) and/or multiple modalities (multispectra). The system performance is measured by the recognition accuracy (AC; the higher the better) and false accept rate (FAR; the lower the better). Specifically, a fusion method will combine the face scores from three matchers (Circular Gaussian Filter, FPB, EBGM) and from two-spectral bands (visible and thermal). We present and compare the system performance using seven fusion methods: linear discriminant analysis (LDA), k-nearest neighbor (KNN), support vector machine (SVM), binomial logistic regression (BLR), Gaussian mixture model (GMM), artificial neural network (ANN), and hidden Markov model (HMM). Our experiments are conducted with the Alcon State University multispectral face dataset that currently consists of two spectral images from 105 subjects. The experimental results show that the KNN score fusion produces the best performance (AC = 98.98%; FAR = 0.35%); and the SVM yields the second best. Compared with the performance of the single best matcher (AC = 91.67%, FAR = 8.33%), the integrated system with score fusion highly improves the accuracy, meanwhile dramatically reduces the FAR.

The goal of the proposed research is to automate the vehicle threat detection w ith X-ray images when a vehicle crosses the country border or the gateway of a secured facility (military base). The proposed detection system requires two inputs: probe images (from X-ray machine) and gallery images (from database). For each vehicle, the gallery images include the X-ray images of fully-loaded (with typical cargo) and unloaded (empty) vehicle. The proposed system produces two types of outputs for threat detection: the detected anomalies and the synthesized images (e.g., grayscale fusion, color fusion, and differential images). The anomalies are automatica lly detected with the block-wise correlation analysis between two temporally aligned images (probe versus gallery). The locations of detected anomalies can be marked with small rectangles on the probe X-ray images. The several side-view images can be combined into one fused image in gray scale and in colors (color fusion) that provides more comprehensive information to the operator. The fused images are suitable for human analysis and decision. We analyzed a set of vehicle X-ray images, which consists of 4 images generated from AS&E OmniView Gantry

Current multispectral night vision (NV) colorization techniques can manipulate images to produce colorized images that closely resemble natural scenes. The colorized NV images can enhance human perception by improving observer object classification and reaction times especially for low light conditions. This paper focuses on the qualitative (subjective) evaluations and comparisons of six NV colorization methods. The multispectral images include visible (Red-Green-Blue), near infrared (NIR), and long wave infrared (LWIR) images. The six colorization methods are channel-based color fusion (CBCF), statistic matching (SM), histogram matching (HM), joint-histogram matching (JHM), statistic matching then joint-histogram matching (SM-JHM), and the lookup table (LUT). Four categries of quality measurements are used for the qualitative evaluations, which are contrast, detail, colorfulness, and overall quality. The score of each measurement is rated from 1 to 3 scale to represent low, average, and high quality, respectively. Specifically, high contrast (of rated score 3) means an adequate level of brightness and contrast. The high detail represents high clarity of detailed conten ts while maintaining low artifacts. The high colorfulness preserves more natural colors (i.e., closely resembles the daylight image). Overall quality is determined from the NV image compared to the reference image. Nine sets of multispectral NV images were used in our experiments. For each set, the six colorized NV images (produced from NIR and LWIR images) are concurrently presented to users along with the reference color (RGB) image (taken at daytime). A total of 67 subjects passed a screening test (“Ishihara Color Blindness Test”) and were asked to evaluate the 9-set colorized images. The experimental results showed the quality order of colorization methods from the best to the worst: CBCF > SM > SM-JHM > LU T > JHM > HM. It is anticipat ed that this work will provide a benchmark for NV colorization and for quantitative evaluation using an objective metric such as objective evaluation index (OEI). Keywords : Color fusion; color mapping; night vision colorization; qualitative evaluation.

Atmospheric clouds are commonly encountered phenomena affecting visual tracking from air-borne or space-borne sensors. Generally clouds are difficult to detect and extract because they are complex in shape and interact with sunlight in a complex fashion. In this paper, we propose a clustering game theoretic image segmentation based approach to identify, extract, and patch clouds. In our framework, the first step is to decompose a given image containing clouds. The problem of image segmentation is considered as a “clustering game”. Within this context, the notion of a cluster is equivalent to a classical equilibrium concept from game theory, as the game equilibrium reflects both the internal and external (e.g., two-player) cluster conditions. To obtain the evolutionary stable strategies, we explore three evolutionary dynamics: fictitious play, replicator dynamics, and infection and immunization dynamics (InImDyn). Secondly, we use the boundary and shape features to refine the cloud segments. This step can lower the false alarm rate. In the third step, we remove the detected clouds and patch the empty spots by performing background recovery. We demonstrate our cloud detection framework on a video clip provides supportive results.

Image fusion has been extensively studied in past two decades. By image fusion algorithms, a composite image (i.e., fused image) can be formed with several images from different sensors. The performance of image fusion methods can be assessed using subjective and/or objective measures. However, subjective evaluation involves human subjects, which significantly increases the cost of time and resource. In this paper, we will discuss objective evaluations of color image fusion algorithms. Given a reference color image and fused color images, we first convert the images into CIELab color space. Then we define four image metrics in CIELab space: the phase congruency metric (PCM), the image gradient magnitude metric (IGMM), the image contrast metric (ICM), and the color natural metric (CNM). Finally, with the four metrics, we propose an objective evaluation index (OEI) for a fused image to measure its similarity with the reference image. The larger the OEI value of a fused image is, the more similar the fused image is with the reference image. To validate the proposed metric, first the fused images are formed with different color fusion algorithms using a set of multispectral images (including visible color images, near infrared images, and long wave infrared images); and then the OEIs of fused images are calculated and compared. Experimental results show that the proposed objective evaluation index is very promising and fits well to subjective evaluation.

We introduced human sciences as components, and integrated them together as an interdisciplinary endeavor over decades. This year, we built a website to maintain systematically the educational research service. We captured the human sciences in various components in the SPIE proceedings over the last decades, which included: (i) ears & eyes like adaptive wavelets, (ii) brain-like unsupervised learning independent component analysis (ICA); (iii) compressive sampling spatiotemporal sparse information processing, (iv) nanoengineering approach to sensing components, (v) systems biology measurements, and (vi) biomedical wellness applications. In order to serve the interdisciplinary community better, our system approach is based on that the former recipients invited the next recipients to deliver their review talks and panel discussions. Since only the former recipients of each component can lead the nomination committees and make the final selections, we also create a leadership award which may be nominated by any conference attendance, to be approved by the conference organization committee.© (2012) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.

A fused image using multispectral images can increase the reliability of interpretation because it combines the complimentary information apparent in multispectral images. While a color image can be easily interpreted by human users (for visual analysis), and thus improves observer performance and reaction times. We propose a fast color fusion method, termed as channel-based color fusion, which is efficient for real time applications. Notice that the term of "color fusion" means combing multispectral images into a color-version image with the purpose of resembling natural scenes. On the other hand, false coloring technique usually has no intention of resembling natural scenery. The framework of channel-based color fusion is as follows, (1) prepare for color fusion by preprocessing, image registration and fusion; (2) form a color fusion image by properly assigning multispectral images to red, green, and blue channels; (3) fuse multispectral images (gray fusion) using a wavelet-based fusion algorithm; and (4) replace the value component of color fusion in HSV color space with the gray-fusion image, and finally transform back to RGB space. In night vision imaging, there may be two or several bands of images available, for example, visible (RGB), image intensified (II), near infrared (NIR), medium wave infrared (MWIR), long wave infrared (LWIR). The proposed channel-wise color fusions were tested with two-band (e.g., NIR + LWIR, II + LWIR, RGB + LWIR) or three-band (e.g., RGB + NIR + LWIR) multispectral images. Experimental results show that the colors in the fused images by the proposed method are vivid and comparable with that of the segmentation-based colorization. The processing speed of new method is much faster than any segmentation-based method.

Image segmentation decomposes a given image into segments, i.e. regions containing "similar" pixels, that aids computer vision applications such as face, medical, and fingerprint recognition as well as scene characterization. Effective segmentation requires domain knowledge or strategies for object designation as no universal segmentation algorithm exists. In this paper, we propose a similarity based image segmentation approach based on game theory methods. The essential idea behind our approach is that the similarity based clustering problem can be considered as a "clustering game". Within this context, the notion of a cluster turns out to be equivalent to a classical equilibrium concept from game theory, as the game equilibrium reflects both the internal and external cluster conditions. We also show that there exists a correspondence between these equilibriums and the local solutions of a polynomial, linearly-constrained, optimization problem, and provide an algorithm for finding the equalibirums. Experiments on image segmentation problems show the superiority of the proposed clustering game image segmentation (CGIS) approach using a common data set of visual images in autonomy, speed, and efficiency. Keywords: Image segmentation, game theory, evolutionary dynamics, fictitious play, replicator dynamics.

There are strong evidences of that multimodal biometric score fusion can significantly improve human identification performance. Score level fusion usually involves score normalization, score fusion, and fusion decision. There are several types of score fusion methods, direct combination of fusion scores, classifier-based fusion, and density-based fusion. The real applications require achieving greater reliability in determining or verifying person's identity. The goal of this research is to improve the accuracy and robustness of human identification by using multimodal biometrics score fusion. The accuracy means high verification rate if tested on a closed dataset, or a high genuine accept rate under low false accept rate if tested on an open dataset. While the robustness means the fusion performance is stable with variant biometric scores. We propose a hidden Markov model (HMM) for multiple score fusion, where the biometric scores include multimodal scores and multi-matcher scores. The state probability density functions in a HHM model are estimated by Gaussian mixture model. The proposed HMM model for multiple score fusion is accurate for identification, flexible and reliable with biometrics. The proposed HMM method are tested on three NIST-BSSR1 multimodal databases and on three face-score databases. The results show the HMM method is an excellent and reliable score fusion method.

A novel orientation code is proposed for face recognition applications in this paper. Gabor wavelet transform is a common tool for orientation analysis in a 2D image; whereas Hamming distance is an efficient distance measurement for multiple classifications such as face identification. Specifically, at each frequency band, an index number representing the strongest orientational response is selected, and then encoded in binary format to favor the Hamming distance calculation. Multiple-band orientation codes are then organized into a face pattern byte (FPB) by using order statistics. With the FPB, Hamming distances are calculated and compared to achieve face identification. The FPB has the dimensionality of 8 bits per pixel and its performance will be compared to that of FPW (face pattern word, 32 bits per pixel). The dimensionality of FPB can be further reduced down to 4 bits per pixel, called face pattern nibble (FPN). Experimental results with visible and thermal face databases show that the proposed orientation code for face recognition is very promising in contrast with classical methods such as PCA.

Multiscale fusion algorithms (wavelet or pyramid) can generally satisfy multisensory image fusion. However, those algorithms are not ideal to fuse visible images and infrared images whose intensities appear inverted. Therefore, a novel orientation-based fusion algorithm is proposed in this paper to address this problem. Specifically, a set of MxN Gabor wavelet transforms (GWT) are performed with two input images (I A and I B ). At each frequency band ( b = 1, 2, ..., M), the index of maximal GWT magnitude between two images is selected pixel by pixel; and then two index frequencies, H A ( b ) and H B ( b ), are calculated as its index accumulation along N orientations, respectively. The final H A and H B are the weighted summations through M bands, where the band weights (W b ) are given empirically. Eventually, the fused image is computed as I F = (I A .* H A + IB .* H B )/( H A + H B ), where '.*' denotes element-by-element product of two arrays. The orientation-based fusion algorithm can be further varied by either keeping DC (direct current) or suppressing DC in GWT. "Keeping DC" will produce a contrast-smooth image; while "suppressing DC" will result a sharpened fusion. Color fusion is achieved by replacing the red channel of a color image with the fused image, which is suitable for poorly illuminated color images. Not only are the fused images of visible and infrared images satisfied, but the fusions of other image sets are also comparable to the results of multiscale fusion algorithms. The proposed algorithm can be applied to multiple (more than two) image fusion.

A reliable thermal face recognition system can enhance the national security applications such as prevention against terrorism, surveillance, monitoring and tracking, especially at nighttime. The system can be applied at airports, customs or high-alert facilities (e.g., nuclear power plant) for 24 hours a day. In this paper, we propose a novel face recognition approach utilizing thermal (long wave infrared) face images that can automatically identify a subject at both daytime and nighttime. With a properly acquired thermal image (as a query image) in monitoring zone, the following processes will be employed: normalization and denoising, face detection, face alignment, face masking, Gabor wavelet transform, face pattern words (FPWs) creation, face identification by similarity measure (Hamming distance). If eyeglasses are present on a subject's face, an eyeglasses mask will be automatically extracted from the querying face image, and then masked with all comparing FPWs (no more transforms). A high identification rate (97.44% with Top-1 match) has been achieved upon our preliminary face dataset (of 39 subjects) from the proposed approach regardless operating time and glasses-wearing condition.e

Human being can perceive natural scenes very well under various illumination conditions. Partial reasons are due to the contrast enhancement of center/surround networks and opponent analysis on the human retina. In this paper, we propose an image enhancement model to simulate the color processes in the human retina. Specifically, there are two center/surround layers, bipolar/horizontal and ganglion/amacrine; and four color opponents, red (R), green (G), blue (B), and yellow (Y). The central cell (bipolar or ganglion) takes the surrounding information from one or several horizontal or amacrine cells; and bipolar and ganglion both have ON and OFF sub-types. For example, a +R/-G bipolar (red-center- ON/green-surround-OFF) will be excited if only the center is illuminated, or inhibited if only the surroundings (bipolars) are illuminated, or stay neutral if both center and surroundings are illuminated. Likewise, other two color opponents with ON-center/OFF-surround, +G/-R and +B/-Y, follow the same rules. The yellow (Y) channel can be obtained by averaging red and green channels. On the other hand, OFF-center/ON-surround bipolars (i.e., -R/+G and -G/+R, but no - B/+Y) are inhibited when the center is illuminated. An ON-bipolar (or OFF-bipolar) only transfers signals to an ONganglion (or OFF-ganglion), where amacrines provide surrounding information. Ganglion cells have strong spatiotemporal responses to moving objects. In our proposed enhancement model, the surrounding information is obtained using weighted average of neighborhood; excited or inhibited can be implemented with pixel intensity increase or decrease according to a linear or nonlinear response; and center/surround excitations are decided by comparing their intensities. A difference of Gaussian (DOG) model is used to simulate the ganglion differential response. Experimental results using natural scenery pictures proved that, the proposed image enhancement model by simulating the two-layer center/surrounding retinal networks can effectively enhance color images in terms of color contrast and image details.

A Gaussian model-based statistical matching procedure is proposed for image enhancement and segmentation. Generally speaking, enhanced images are desired for visual analysis whereas segmented images are required for target recognition. A histogram matching procedure is used to enhance a given image. To perform histogram matching, two histograms are needed, an original histogram computed from the given image and a specified histogram to be matched to. For image enhancement, the specified histogram is a Gaussian model (mean & standard deviation) that can be estimated from a number of well-exposed images or properly processed images. Certainly the Gaussian model varies with the category of imagery. For image segmentation, N Gaussian models (means & standard deviations) are estimated from the original histogram of a given image. The number of Gaussian models (N) is decided by analyzing the original histogram. A statistical matching procedure is used to map the original histogram onto one of the Gaussian models defined by their means and standard deviations. Specifically, the mapped image can be computed by subtracting the mean of original image from the original image, scaling with the ratio of the standard deviation of Gaussian model to the standard deviation of original image and plus the mean of Gaussian model. The statistically mapped image is thresheld by using the mean of Gaussian model, which results one set of expected segments. The statistical matching plus thresholding procedure is repeated N times for N Gaussian models. Finally, all N sets of segments are fully obtained. The proposed image enhancement and segmentation procedure are validated with multi-sensor imagery.

Breast cancer is the leading cancer among American women. The current lifetime risk of developing breast cancer is 13.4% (one in seven). Mammography is the most effective technology presently available for breast cancer screening. With digital mammograms computer-aided detection (CAD) has proven to be a useful tool for radiologists. In this paper, we focus on mass detection that is a common category of breast cancers relative to calcification and architecture distortion. We propose a new mass detection algorithm utilizing Gabor filters, termed as "Gabor Mass Detection" (GMD). There are three steps in the GMD algorithm, (1) preprocessing, (2) generating alarms and (3) classification (reducing false alarms). Down-sampling, quantization, denoising and enhancement are done in the preprocessing step. Then a total of 30 Gabor filtered images (along 6 bands by 5 orientations) are produced. Alarm segments are generated by thresholding four Gabor images of full orientations (Stage-I classification) with image-dependent thresholds computed via histogram analysis. Next a set of edge histogram descriptors (EHD) are extracted from 24 Gabor images (6 by 4) that will be used for Stage-II classification. After clustering EHD features with fuzzy C-means clustering method, a k-nearest neighbor classifier is used to reduce the number of false alarms. We initially analyzed 431 digitized mammograms (159 normal images vs. 272 cancerous images, from the DDSM project, University of South Florida) with the proposed GMD algorithm. And a ten-fold cross validation was used for testing the GMD algorithm upon the available data. The GMD performance is as follows: sensitivity (true positive rate) = 0.88 at false positives per image (FPI) = 1.25, and the area under the ROC curve = 0.83. The overall performance of the GMD algorithm is satisfactory and the accuracy of locating masses (highlighting the boundaries of suspicious areas) is relatively high. Furthermore, the GMD algorithm can successfully detect early-stage (with small values of Assessment & low Subtlety) malignant masses. In addition, Gabor filtered images are used in both stages of classifications, which greatly simplifies the GMD algorithm.

In magnetic resonance imaging (MRI), there are three bands of images (“MRI triplet”) availa ble, which are T1-, T2- and PD-weighted images. The three images of a MRI triplet provide complementary st ructure information and therefore it is useful for diagnosis and subsequent analysis to combine three-band images into one. We propose an advanced discrete wavelet transform ( aDWT) for three-band MR I image fusion and the aDWT algorithm is further optimized utilizing two quantitative fusion metrics – the image quality index (IQI) and ratio spatial frequency error (rSFe). In the aDWT method, principle component analysis (PCA) and morphological processing are incorporated into a regular DWT fusion algorithm. Furthermore, the aDWT has two adjustable parameters – the level of DWT decomposition (L

A natural color mapping method has been previously proposed that matches the statistical properties (mean and standard deviation) of night-vision (NV) imagery to those of a daylight color image (manually selected as the "target" color distribution). Thus the rendered NV image appears to resemble the target image in terms of colors. However, in this prior method the colored NV image may appear unnatural if the target image's "global" color statistics are too different from that of the night vision scene (e.g., it would appear to have too much green if much more foliage was contained in the target image). Consequently, a new "local coloring" method is presented in the current paper, and functions to render the NV image segment-by-segment by using a histogram matching technique. Specifically, a false-color image (source image) is formed by assigning multi-band NV images to three RGB (red, green and blue) channels. A nonlinear diffusion filter is then applied to the false-colored image to reduce the number of colors. The final grayscale image segments are obtained by using clustering and merging techniques. The statistical matching procedure is merged with the histogram matching procedure to assure that the source image more closely resembles the target image with respect to color. Instead of using a single target color image, the mean, standard deviation and histogram distribution of a set of natural scene images are used as the target color properties for each color scheme. Corresponding to the source region segments, the target color schemes are grouped by their scene contents (or colors) such as green plants, roads, ground/earth. In our experiments, five pairs of night-vision images were initially analyzed, and the images that were colored (segment-by-segment) by the proposed "local coloring" method are shown to be much more natural, realistic, and colorful when compared with those produced by the "global-coloring" method.

There are numerous applications for image fusion, some of which include medical imaging, remote sensing, nighttime operations and multi-spectral imaging. In general, the discrete wavelet transform (DWT) and various pyramids (such as Laplacian, ratio, contrast, gradient and morphological pyramids) are the most common and effective methods. For quantitative evaluation of the quality of fused imagery, the root mean square error (RMSE) is the most suitable measure of quality if there is a “ground truth” image available; otherwise, the entropy, spatial frequency or image quality index of the input images and the fused images can be calculated and compared. Here, after analyzing the pyramids’ performance with the four measures mentioned, an advanced wavelet transform (aDWT) method that incorporates principal component analysis (PCA) and morphological processing into a regular DWT fusion algorithm is presented. Specifically, at each scale of the wavelet transformed images, a principle vector was derived from two input images and then applied to two of the images’ approximation coefficients (i.e., they were fused by using the principal eigenvector). For the detail coefficients (i.e., three quarters of the coefficients), the larger absolute values were chosen and subjected to a neighborhood morphological processing procedure which served to verify the selected pixels by using a “filling” and “cleaning” operation (this operation filled or removed isolated pixels in a 3-by-3 local region). The fusion performance of the advanced DWT (aDWT) method proposed here was compared with six other common methods, and, based on the four quantitative measures, was found to perform the best when tested on the four input image types. Since the different image sources used here varied with respect to intensity, contrast, noise, and intrinsic characteristics, the aDWT is a promising image fusion procedure for inhomogeneous imagery.

Currently, new applications demand utilizing CT scout images for diagnostic purposes. However, many CT scout images cannot be used diagnostically due to their poor resolution, particularly in the direction of table movement. Spatial resolution generally can be improved with image restoration techniques. Based on the principles of Wiener filtering and inverse filtering, this paper presents a modified Wiener filtering approach in the frequency domain. The concept of equivalent target point spread function is introduced, which makes the restoration process steerable. Consequently, balancing resolution improvement with noise suppression is facilitated. Experiments compare the image quality with traditional inverse filtering and Wiener filtering. The modified Wiener filtering method has been shown to restore the scout image with higher resolution and lower noise.