Search

This paper presents a novel Dempster-Shafer evidence construction approach for aircraft aim recognition. The prior-probability of the properties of aircraft was used for establishing a probabilistic argumentation system. Dempster-Shafer evidence was constructed by assumption-based reasoning. Therefore, additional information could be provided to the classification of the data fusion system. Experiments on artificial and real data demonstrated that the proposed method could improve the classification results.

Human vision system exploits this fact by visual selective attention mechanisms towards important and informative regions. A computational model of combination both bottom-up and top-down simulating human vision system for machine vision inspection is proposed. In this model, top-down knowledge-based information is highlighted to integrate into bottom-up stimulus-based process of vision attention. The model is tested on inspecting contaminants in cotton images. Experiment result shows that the proposed model is feasible and effective in visual inspection. And it is available and quasi-equivalent to human vision attention.

Motion blur due to camera shake during exposure is one of the most common reasons of image degradation, which usually reduces the quality of photographs seriously. Based on the statistical properties of the natural image's gradient and the blur kernel, a blind deconvolution algorithm is proposed to restore the motion-blurred image caused by camera shake, adopting the variational Bayesian estimation theory. In addition, the ring effect is one problem that is not avoided in the process of image deconvolution, and usually makes the visual effect of the restored image badly. So a dering method is put forward based on the sub-region detection and fuzzy filter. Tested on the real blurred photographs, the experimental results show that the proposed algorithm of blind image deconvolution can remove the camera-shake motion blur from the degraded image effectively, and can eliminate the ring effect better, while preserve the edges and details of the image well.

Overlap-blur is caused by the relative movement of high speed between the camera and the object during the exposure process, which is one of the most common phenomenons of image degradation during the criminal detection forensics work. Based on the analysis of the overlap-blurred image’s characteristic, a coded-shutter model is proposed to approximate the nature of overlap-blur. As the first attempt, using the coded-shutter model, an image deblurring algorithm is designed for the restoration of the overlap-blurred images. The experiment results show the validity and rationality of the coded-shutter model for deblurring the overlap-blurred images. When tested on the real overlap-blurred photographs, the proposed algorithm can restore the information of interest in the blurred images better, which demonstrates the higher practical value of the algorithm.

Nowadays, digital images can be easily tampered due to the availability of powerful image processing software. As digital cameras continue to replace their analog counterparts, the importance of authenticating digital images, identifying their sources, and detecting forgeries is increasing. Blind image forensics is used to analyze an image in the complete absence of any digital watermark or signature. Image compositing is the most common form of digital tampering. Assuming that image compositing operations affect the inherent statistics of the image, we propose an image compositing detection method on based on a statistical model for natural image in the wavelet transform domain. The generalized Gaussian model (GGD) is employed to describe the marginal distribution of wavelet coefficients of images, and the parameters of GGD are obtained using maximum-likelihood estimator. The statistical features include GGD parameters, prediction error, mean, variance, skewness, and kurtosis at each wavelet detail subband. Then, these feature vectors are used to discriminate between natural images and composite images using support vector machine (SVM). To evaluate the performance of our proposed method, we carried out tests on the Columbia Uncompressed Image Splicing Detection Dataset and another advanced dataset, and achieved a detection accuracy of 92 % and 79 %, respectively. The detection performance of our method is better than that of the method using camera response function on the same dataset.

Image forensics is a form of image analysis for finding out the condition of an image in the complete absence of any digital watermark or signature. It can be used to authenticate digital images and identify their sources. While the technology of exemplar-based inpainting provides an approach to remove objects from an image and play visual tricks. In this paper, as a first attempt, a method based on zero-connectivity feature and fuzzy membership is proposed to discriminate natural images from inpainted images. Firstly, zero-connectivity labeling is applied on block pairs to yield matching degree feature of all blocks in the region of suspicious, then the fuzzy memberships are computed and the tampered regions are identified by a cut set. Experimental results demonstrate the effectiveness of our method in detecting inpainted images.

Crowd collectiveness measuring has attracted a great deal of attentions in recently years. We adopt the path integral descriptor idea to measure the collectiveness of a crowd system. A new path integral descriptor is proposed by exponent generating function to avoid parameter setting. Several good properties of the proposed path integral descriptor are demonstrated in this paper. The proposed path integral descriptor of a set is regard as the collectiveness measure of a set, which can be a moving system such as human crowd, sheep herd and so on. Self-driven particle (SDP) model and the crowd motion database are used to test the ability of the proposed method in measuring collectiveness.

In this paper, we propose an algorithm to address the problem of automatic extraction of visual context in surveillance video scene. Firstly, we analyze the trajectory information distribution to get the characteristic of the information; then by using Gaussian mixture model we do the cluster analysis of the moving tendency reflected by the trajectory direction. For the similar direction trajectory segments, we do the hierarchical clustering based on the proposed similarity measure to get the key routine of the scene. At the end of this paper, the experiments result shows the efficiency of our algorithm.

This paper presents an event detection approach in crowd surveillance videos based on motion vector intersection points. It contains three steps: firstly, to extract the local motion vectors by feature tracking. Secondly, to select appropriate pairs of motion vectors and calculate three types of intersection points which represent the spatial character of crowd event. And the final step is to obtain the intersection point clusters by density based clustering, and then to detect the events by searching the most possible candidate and voting. Experimental results show that the presented approach can effectively detect the concurrent events of different densities and within different ranges controlled by parameters. The results also show that the proposed approach is robust to illumination, shadows and noise from event itself.

Using Behavior Entropy model, we introduce a novel method to detect and localize abnormal behaviors in crowd scenes. Our key insight is to estimate the behavior entropy of each pixel and whole scene by considering defined pixels' behavior certainty. For this purpose, we introduce information theory and energetics concept to define pixel's behavior certainty based on video's spatial-temporal information. Scene entropy behavior and behavior entropy image can be used to detect and localize anomalies respectively. We discuss parameters' setting by analyzing how they influence model's detecting and localizing abilities, and our model is robust to parameter setting. The experiments are conducted on several publicly available datasets, and show that the proposed method captures the dynamics of the crowd behavior successfully. The results of our method, indicates that the method outperforms the state-of-the-art methods in detecting and localizing several kinds of abnormal behaviors in the crowd.

If a camera moves fast in low light environments, large scale motion blur is induced. Algorithms for deblurring motion blur images mostly fail miserably when applied to large scale motion blur images. The motion path of the camera is complex, and the estimation of accurate point spread function(PSF) for a large scale motion blur image is always a difficult problem. Firstly, a hybrid camera which is composed by a high resolution camera and a high speed assist camera is used for deblurring large scale motion blur images. Then, an algorithm is proposed to compute the accurate PSF using the displacements. Finally, Richardson-Lucy algorithm is used for deblurring the high resolution motion blur images. The experiments show that this kind of technique can deblur large scale motion blur images efficaciously and the details of the scene such as characters can be recognized by deblurred images.

In recent years, research of human vision attention opens a new field for machine vision. A novel and simplified model including saccade and gaze is presented in this paper, which simulates human vision attention. An efficient algorithm is proposed for extracting simple features in global scene and extracting candidate features in local saliency region so as to reduce computational cost. This simplified method is very reliable in some cases of machine vision system and is nearly equivalent to that obtained from human vision attention.

This paper studies the problem of mining Chinese comparative sentences in text documents by using semantic role labeling (SRL). The comparative opinion can be divided into six semantic roles: holder, entity 1, comparative predicates, entity 2, attributes and sentiments. These six opinion elements were recognized and labeled by using SRL. A corpus of Chinese comparative sentences was manually labeled at first. Then a conditional random fields (CRFs) model was trained by learn from the corpus. Finally new comparative sentences were labeled by using this CRFs model, and comparative relations were extracted afterward.

Exemplar-based inpainting technique can be used to remove objects from an image and play visual tricks, which would affect the authenticity of images. In this paper, a blind detection method based on zero-connectivity feature and fuzzy membership is proposed to detect the specific doctoring. Firstly, zero-connectivity labeling is applied on block pairs to yield matching degree feature for all blocks in the region of suspicious, and fuzzy memberships are computed by constructing ascending semi-trapezoid membership function. Then the tampered regions are identified by a cut set. A num of natural and inpainted forged images are used to show the effectiveness of our method in detecting digital doctoring.

As a critical step in many multimedia applications, shot boundary detection has attracted many research interests in recent years. The most of existing methods measure the similarity among video frames based on its low-level feathers. However, they are sensitive to the change in not only brightness, color, motion of object, but also camera motions and the quality of video. This paper proposes an innovative shot boundary detection method for news video based on video object segmentation and tracking. It combines three main techniques: the partitioned histogram comparison method, the video object segmentation and tracking based on wavelet analysis. The partitioned histogram comparison is used as the first filter to effectively reduce the number of video frames which need object segmentation and tracking. The unsupervised video object segmentation and tracking based on wavelet analysis is robust to those problems mentioned above. The efficacy of the proposed method is extensively tested with more than 3 hours of CCTV and CNN news programs, and that 96.4% recall with 97.2% precision have been achieved.

To discover similar video segments from surveillance video sequence, a new approach is proposed for clustering motion data of moving objects. A simple background subtraction algorithm is used to get the binary mask of moving objects for segmenting the video sequence captured by fixed camera. Then a mixture of hidden Markov models (HMMs) using the expectation-maximization (EM) scheme is fitted to the motion data extracted from the binary mask. Unlike previous literatures using k-means where every observed data set only assigned to a single HMM, the proposed approach allows every video segment to belong to more than a single HMM with some probability. Experiments with real data demonstrate the benefit when there is more "overlap" in the processes generating the data. The promising potential of HMM-based motion clustering for mining similar video segments from surveillance video is also indicated by the experimental results.

With the vast use of video surveillance systems, there are more and more video data. An exciting field called video mining is now putting forward which focuses on extracting semantic info, implicit patterns and knowledge from video data. In this paper, a surveillance video data mining approach is proposed to discover similar video segments from surveillance video through a probabilistic model. First, a simple background subtraction algorithm is utilized to get the binary mask of moving objects. So the motion of every frame is calculated to segment the sequence of surveillance video. Then a mixture of hidden Markov models using the expectation-maximization scheme is fitted to the motion data with some probability to identity the similar segments. Finally, abnormal events and meaningful patterns are mined. Experiments with real-time video demonstrate the promising potential of this approach.

To structuralize video streams plays an important role in the processing of video. The basic structure for video is a hierarchical structure which consists of four kinds of components, namely frame, shot, scene, and video program. A simple framework for video hierarchical structure mining is to partition continuous video frames into discrete physical shots, extract features from video shots and construct scene structure based on shots. In this paper, two crucial algorithms of video hierarchical structure mining, Multi-features Shot Clustering (MSC) and Scene Change Detection (SCD), are proposed based on color, texture and semantic similarity of shot. Our experimental results demonstrate the performance of SCD is better than that of MSC.

To recognize an underwater target precisely is always a very difficult task for the navy due to the interference-filled under sea. Sonar is the most efficient way to detect items in the underwater world but the recognition still depends on sonarman. As well known, the feature extraction method is the key of automatic target recognition. In this paper, a model of 2-dimensional metadata of echo is defined, which is based on echo's frequency and temporal domain information. It contains two features, energy difference and zero cross rate. This paper concentrated on extraction of every feature and the organization method. Experiment results show the effectiveness of the presented approach.

To recognize an underwater target precisely is always a more difficult task for the navy compared to the air force due to the complicated watery environment which is very different from the aerial circumstance. Part of the reason is that there is much more interference under the sea. Sonar is the most efficient way to detect items in the underwater world at the present time. In this paper, a genetic-based classifier system is designed which recognizes targets by sonar fingerprints. This method will, to a certain degree, relieve the sonar man of some of his work. Experiments show that the system gains acceptable speed and accuracy in the classifying operation. The proposed underwater target classifier system is highly automatic, with quite finite hardware requirements for operation.

To access video content, the suitable organization of video data is critical. The basic structure of video is a hierarchical structure that consists of four kinds of components, namely frame, shot, scene, and video program. To achieve such a structure, not only the shot detection and shot similarity, but also the scene structure construction are important. A simple framework for video hierarchical structure mining is to segment continued video frames into discrete physical shots, extract features from video shots and construct scene structure based on shots. In this paper, an efficient shot detection algorithm is proposed, and develop a novel algorithm for measuring video shot similarity is develeped and Scene Change De tection (SCD) is adopt to construct video scene based on color, texture and semantic similarity between shots. Our experimental results demonstrate the performance of the algorithms is promising.

As well known, because of the complicated watery environment and the limitation of the detecting method, recognizing the underwater target precisely is always a Gordian knot to all countries. In this paper, we designed a genetic-based classifier system (CS) which recognizes a target utilizes audio fingerprints. Audio fingerprint technology extracts some unique feature from a given sound clip, and the classifier system does some classifying and recognizing works according to those features. In order to improve the performance of the classifier system, we designed some techniques: the comparing and matching algorithm would give the fitness value more explicitly statistical meaning; the hyperplasia operator gives the system persistent learning abilities, the refining classifier merges redundant rules and shrinks the rule set, and an alterable mutation probability increases the speed and the accuracy of the classifying operation. Experiments show that it's competent for the recognition

This work introduces a fast block-based motion estimation algorithm named two-phase motion estimation algorithm (TPMEA). The idea of the algorithm is to divide motion estimation procedure into two-phase: in the first phase, the sum of pixels in a line (or a column) is used as the 1D vector; In the second phase, a filtering threshold is deduced from the 1D vector matching, as Minkowski inequality has shown, blocks cannot match well if their corresponding 1D vector do not match well. Hence, the expensive 2D block matching need only be performed in a limited scope, which saves a lot of time. Moreover, an efficient implementation to calculate 1D vector is presented to speed up encoder further. From the experiment conducted in the test model of H.264(version:JM61d), the algorithm presented in This work can save approximately 30% of the time compared with the exhaustive search algorithm and achieve the same image quality. It can satisfy the demand for high performance and real-time video encoding.

To recognize the underwater target precisely is always a hard problem to all countries. In this paper, we designed a genetic-based classifier system which recognizes targets utilize sonar fingerprints. Exceptionally some improvements have been designed on it. The proposed Comparing and Matching algorithm would give the fitness value more explicitly statistical meaning, which would make user easier to explain the rules with background knowledge. The proposed hyperplasia operator can handle those instances which were not emerged before. It gives the system persistent learning abilities, so the system may be more compatible with the surroundings. The proposed refining classifier merges redundant rules and shrinks the rule set In addition, an alterable mutation probability is set in the genetic algorithm, experiment shows that this strategy increased the speed and the accuracy of the classifying operation. Sonar fingerprint technology extracts unique feature from an echo, it is similar to that one's fingerprint can identify a unique person himself, dissimilar echo leads to different fingerprint. And all these have none business with the echo's store way (analog or digital) or format (WAV, MP3, WMA, RM, and etc). The proposed underwater target classifier system is highly automatic, with quite finite hardware requirements in operating.

Due to the complicated watery environment and the limitation of the detecting method to the underwater target, recognizing them precisely is always a hard problem to all countries. In this paper, we design a genetic-based classifier system (CS) and apply it to recognize the underwater target. This is an attempt to solve this problem with machine learning way. Compared with traditional CS, the proposed Comparing and Matching Algorithm will give the fitness value more explicit statistical meaning, which will make us easier to explain the rules with background knowledge. The proposed Hyperplasia Operator can handle those instances which are not emerged before. It gives the system persistent learning abilities, so the system may be more compatible with the surroundings. The proposed Refining Classifier merges those redundant rules and shrinks the rule set. In addition, we give and discuss an alterable mutation probability to the genetic algorithm in the CS, which increases the speed and the accuracy of the classifying operation. At last, the experiments to examine the system with the sample data which are collected from sonar echo samples. Experiments show that the recognizing results are satisfying.