Search

This paper proposed an unsupervised change detection method for water body extraction and change detection with multi-temporal SAR images. Firstly, two optimal thresholds are estimated according to the strategy of maximum mutual information, in which computation efficiency is largely improved based on integral image. Secondly, water body extraction is done simultaneously in both input images by optimal thresholds. Finally, by fusing of two segmented results, change detection can be achieved. Experimental results demonstrate the effectiveness of the proposed approach.

In this paper, we investigate object tracking in video sequences and propose a particle filter based tracking algorithm with color and texture information fusion, in which the target model is jointly represented by spatial-weighted color histogram and LBP (Local Binary Patterns) texture histogram. The property of local grayscale or color invariance for LBP operator makes it more reliable to measure the spatial structure of local image texture. The system is less sensitive to illumination changes and partial occlusions, and can be able to track objects in diverse conditions. Experimental results demonstrate that the performance of the proposed method is more robust and accurate than the original color based method, especially when tracking objects with similar color appearance to the background and partial occlusions.

This paper presents a novel fusion approach using PCA merger based on multiscale decomposition (MSD), combined with region segmentation and support vector machine (SVM), the result is a high spatial resolution multispectral image from a high resolution panchromatic (Pan) image and low resolution multispectral (Ms) images. Principal components analysis (PCA) fusion technique is one of typical fusion methods, and PCA merger based on MSD had been proposed which can obtain better performance. As we know that, in pixel fusion level, the original images are fused as internal region regardless of the contents of images, but in this paper, we perform region segmentation after MSD, because the homogeneous regions have similar features such as color, texture and intensity. Traditionally, various fusion rules can be applied after MSD according to different conditions, however, the crucial problem is which fusion rule should be adopted under given condition, hence we use the SVM to combine the most fusion rules so that can avoid some drawbacks using single fusion rule. To validate our approach, we compare it with several typical fusion approaches, and the best result is obtained using our approach.

In this paper, we present a novel unsupervised change detection approach in temporal sets of synthetic aperture radar (SAR) images using Markovian fusion. This method is carried out within a Markovian framework which combines two different change detection algorithms to achieve noise removing and spatial information preserving at the same time. This approach is composed of two steps: 1) two change maps are generated by two distinctive but complementary approaches respectively; 2) final results are achieved by fusing the two change maps within a Markovian framework. In the first step, two different thresholding algorithms are selected to get two change maps aimed at speckle noise removing and spatial contexture preserving respectively; In the second step, a solution to fusion the two change maps through a Markov random field framework is proposed. The minimization of energy function is carried out through iterative conditional mode (ICM) algorithm because of its simplicity and moderate computation-consuming. Experiments results obtained on a SAR data set confirm the effectiveness of the proposed approach. It shows that the fusion approach based on MRFs model is a promising way of achieving robust unsupervised change detection.

Remote sensing image registration is the key issue for change detection. To reduce the effects of misregistration on the accuracy of change detection, a hybrid registration method is proposed in this paper. First, we use the registration approach based on SIFT(Scale Invariant Feature Transform) to get the initial parameters, and then area-based method is employed to refine the performance of registration. In order to improve the efficiency of computation, the multiresolution based coarse-to-fine strategy is adopted during the refined procedure. In contrast with feature-based or area-based method, our hybrid method is accurate, robust and automated since it integrates the merits of both approaches. The experiments on simulated and real images show the promising performance of the proposed method.

This paper proposes a new method to denoise images corrupted by Poisson noise. Poisson noise is signal-dependent, and consequently, separating signals from noise is a very difficult task. In most current Poisson noise reduction algorithms, noise signal are pre-processed to approximate Gaussian noise, and then denoised by a conventional Gaussian denoising algorithm. In this paper, we propose to use adaptive basis functions derived from the data using modified ICA (Independent Component Analysis), and a maximum likelihood shrinkage algorithm based on the property of Poisson noise. This modified ICA method is based on a denoising method called "Sparse Code Shrinkage (SCS)" and wavelet-domain denoising. In denoising procedure of ICA-domain, the shrinkage function is determined by the property of Poisson noise that adapts to the intensity of signal. The performance of the proposed algorithm is validated with simulated data experiments, and the results demonstrate that the algorithm greatly improves the denoising performance in images contaminated by Poisson noise.

This paper present an approach based on improved hybrid elastic model for deformable image registration. This method utilizes the linear spring net model for correspondence, and the thin-plate spline for non-rigid mapping. Compared with the original model, this improved method uses not only the intensity information, but also the weighted gray level histogram and the image gradient for similarity measurement. A multi-resolution strategy is involved to decrease the computing complexity and to increase the precision. Some experiments are performed on both synthetic and segmented medical images.

In this article we propose a hierarchical partition of the feature space based on statistical information on each dimension, and then use mean-shift to properly fuse the obtained super-cubics to reveal the genuine data structure. It not only greatly reduces calculation, but also provides a desirable priori knowledge for bandwidth selection.

Image segmentation is one of the most challenging problems in image processing. While significant progress has been made in gray-scale texture segmentation and color segmentation problem separately, the combined color and texture segmentation problem is less considered. In this paper, we use independent component analysis to extract local color and texture features for segmentation. Experiments compared with gray-scale texture analysis method show that the proposed method is more effective in segmenting complex color and texture images.

In this paper, we propose a robust approach for object tracking in infrared imagery. Our method mainly applies the image intensity histogram distribution and intensity projection distributions and computes a likelihood measure between the candidate and the model distributions by evaluating the Mean Shift Vector. In addition, Gabor filters are applied here to enhance the contrast of the object with the background, and then the scale of the track window can be selected according to the variable object size. Our method greatly improves the accuracy of object tracking and can update the model frame by frame, which means the object model does not necessarily depend on that of the first frame. The robustness of our method is supported by several different infrared imagery sequences.

Face technologies which can be applied to access control and surveillance, are essential to intelligent vision-based human computer interaction. The research efforts in this field include face detecting, face recognition, face retrieval, etc. However, these tasks are challenging because of variability in view point, lighting, pose and expression of human faces. The ideal face representation should consider the variability so as to we can develop robust algorithms for our applications. Independent Component Analysis (ICA) as an unsupervised learning technique has been used to find such a representation and obtained good performances in some applications. In the first part of this paper, we depict the models of ICA and its extensions: Independent Subspace Analysis (ISA) and Topographic ICA (TICA).Then we summaraize the process in the applications of ICA and its extension in Face images. At last we propose a promising direction for future research.

This paper presents a new method of non-rigid registration based on elastic models incorporating shape information. Given a source and a target image, and the shape of interest in these two images, the idea is to incorporate the geometry-based shape knowledge into extemal forces, which will drive the deformation from the source image to the target image. The transformations are constrained to a physical model of elasticity to maintain smoothness and continuity. Some experiments were performed on both synthetic and medical images of brain. Keywords: non-rigid registration, physical, elastic model, deformation, geometry information, medical image 1. INTRODUCTION According to Toga [7], non-rigid registration is a process for maximizing spatial image correspondence of twoimages within the constraints of an image transformation model, in order to bring the features of first image intoalignment with those of the second image. In many application fields, especially in medical image, detailed anatomicalstructure differs between the image pair to be registered; these differences can be quite huge. The precise quantificationof structure changes is important for comparing function or morphology between individuals. The increasing needs for

Relevance feedback has been proven to be an effective scheme in content-based image retrieval to improve retrieval performance. Recently, SVM based techniques are introduced into the learning process of relevance feedback, for its good generalization ability in a high dimensional space in condition of small example size. Based on the extended one-class SVM that can handle negative examples, we propose a new relevance feedback scheme to overcome the limitation of the recent methods. The scheme is flexible to handle the situations with and without negative examples; and can further improve the retrieval performance when negative examples provided. The new scheme was evaluated on a database of 6,000 images and compared to previous methods. Experimental results have demonstrated its effectiveness.

Virtual Endoscopy System is a new aided diagnosis method based on computer processing of 3D image slices to provide simulated visualizations of specific organs similar to those produced by standard endoscopy. Compare with real endoscopy, VES has much advantages and will have more applications in the future. We constructed a Virtual Bronchus Endoscopy System based on the techniques of image analysis, compute graphics, and so on. Based on the characteristic of bronchus, we adopted an improved 3D region-growing algorithm, which we call 3D scanline algorithm to extract the bronchus from the DICOM-formatted medical images, then the 3D polyhedral surface model of bronchus is obtained by triangulation with Marching Cubes Algorithm. Then the user is allowed to navigate freely inside the bronchus along the axis. We adopted surface rendering method in the rendering process. In application this system can meet the requirement of real-time navigation and has pretty good display.

In this paper we present an automated region-growing algorithm designed to extract the pipe-like organs, i.e. bronchus, colon and blood vessel, from 3D CT images taken by helical CT scanner. The algorithm is called 3D scanline- filling algorithm, which is a kind of region-growing. This algorithm is very fast, and its parameters can be adjusted automatically. It can be utilized in Virtual Endoscopy system (VES), which is an important computer-aided diagnosis method. The data created after applying this algorithm can then be sent to VES for further process, be visualized and displayed on screen. The physician can observed the inner organ on the screen just like with a real endoscopy without any pain.© (2001) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.