Target Localization,Detection And Tracking Based On Local Self-Similarity Descriptor

We receive large amount of visual information through our eyes.Our brain can filter most of the information automatically,so the region of interest can be noticed.Target objects are usually located in these salient regions.In pattern recognition research area,there is a need to use a compact vector to represent local region.In this way,the pattern of an object or a patch can be preserved.Then we can search the target object in search image with the vector or descriptor.The local self-similarity descriptor can acquire the local layout pattern and is invariant to small rotation and transformation.Our research focus on the target object localization,detection,tracking algorithms based on local self-similarity descriptor.First we analyze its performance when faced with rotation scenes.Then local self-similarity descriptor is used to represent points for points matching between images with different viewpoints.After reconstruction of several points of interest,the target position can be localized.As targets always locate in salient regions of an image,the performance of detection algorithm based on saliency and local self-similarity is studied.The salient regions are extracted first,and then the target is searched in these regions by using local self-similarity matching.The effectiveness of this detection algorithm is demonstrated in this paper.We also use local self-similarity descriptor to help handle the occlusion and deformation problems in tracking.The descriptor is used to represent the target,as well as the positive and negative samples.The performance of tracking algorithm is promoted according to the experiment result.This paper demonstrates the effectiveness of layout pattern acquired by the local self-similarity descriptor through the experiments in several visual research areas.In most pattern recognition algorithms,the extraction of feature descriptor is an indispensable procedure.So this study on the local self-similarity descriptor is of great meaning.