نشریه فناوری اطلاعات و ارتباطات ایران
سال چهارم شماره 13 (پاییز و زمستان 1391)

The mean shift algorithm is one of the popular methods in visual tracking for non-rigid moving targets. Basically، it is able to locate repeatedly the central mode of a desirable target. Object representation in mean shift algorithm is based on its feature histogram within a non-oriented individual kernel mask. Truly، adjusting of the kernel scale is the most critical challenge in this method. Up to now، no methods are presented that can perfectly as well as efficiently adjust and adapt the kernel scale during track when a target is resized. Another problem of mean shift tracking algorithm will be encountered whenever photometric properties of target texture changes. In order to solve these problems، this paper presents a modified mean shift tracking algorithm that is used a robust adaptive sizing technique. It can also cope with photometric changes of target template by adapting of its model in every frame of image sequence. In our proposed method، at first، the target window is adaptively resized with respect to spatio-temporal gradient powers of its pixel intensities in current frame and then mean shift algorithm is consequently applied to the resulted sizing window. Compared to standard mean shift algorithm، experimental results show that our proposed method، not only reduces center location errors of target، but also efficiently track it in the presence of changing illumination.

Hyper spectral remote sensing imagery، due to its rich source of spectral information provides an efficient tool for ground classifications in complex geographical areas with similar classes. Referring to robustness of Support Vector Machines (SVMs) in high dimensional space، they are efficient tool for classification of hyper spectral imagery. However، there are two optimization issues which strongly effect on the SVMs performance: Optimum SVMs parameters determination and optimum feature subset selection. Traditional optimization algorithms are appropriate in limited search space but they usually trap in local optimum in high dimensional space، therefore it is inevitable to apply meta-heuristic optimization algorithms such as Genetic Algorithm to obtain global optimum solution. This paper evaluates the potential of different proposed optimization scenarios in determining of SVMs parameters and feature subset selection based on Genetic Algorithm (GA). Obtained results on AVIRIS Hyper spectral imagery demonstrate superior performance of SVMs achieved by simultaneously optimization of SVMs parameters and input feature subset. In Gaussian and Polynomial kernels، the classification accuracy improves by about 5% and15% respectively and more than 90 redundant bands are eliminated. For comparison، the evaluation is also performed by applying it to Simulated Annealing (SA) that shows a better performance of Genetic Algorithm especially in complex search space where parameter determination and feature selection are solve simultaneously.

I In this paper، a new filter is designed to enhance medial-axis of tubular structures. Based on a multi-scale method and using eigenvectors of Hessian matrix، the distance of a point to the edges of the tube is found. To do this، a hypothetical line with a deliberate direction is passed through the point which cuts the tube at its edges. For points which are located on the medial-axis، this distance is symmetric with respect to any deliberate direction. We find samples of the distances in different directions and assign a measure to the points based on this symmetry property. The output of this step is an enhanced image in which noise is removed and tubes can be seen more clearly. Then، we employ the filter developed by Pock et al. to enhance medial axis. Evaluation of the proposed method is performed using 2D/3D synthetic/clinical datasets both quantitatively and qualitatively.

In different steps of steel production، various defects appear on the surface of the sheet. Putting aside the causes of defects، precise identification of their kinds helps classify steel sheet correctly، thereby it allocates a high percentage of quality control process. QC of steel sheet for promotion of product quality and maintaining the competitive market is of great importance. In this paper، in addition to quick review of image process techniques used، using image process by means of Gabor wavelet، a fast and precise solution for detection of texture defects in steel sheet is presented. In first step، the approach extracts considerable texture specification from image by using Gabor wavelet. The specification includes both different directions and different frequencies. Then using statistical methods، images are selected that have more obvious defects، and location of defects is determined. Offering the experimental samples، the accuracy and speed of the method is indicated.