The Color Texture Image Classification And Target Tracking Based On HSV

In recent years,color texture image classification and target tracking is one of the hot spots in the field of computer vision.At present,the classification of color texture image still has some problems,such as weak robustness to illumination change,weak anti-noise ability and excessive feature dimension.The weak anti-occlusion ability is one of the problems in target tracking,which will lead to drift in the tracking process.Aiming at the above problems,this paper carries out the research.In view of the problems existing in some texture image classification algorithms,such as weak robustness to illumination change,weak anti-noise ability and excessive feature dimension,the algorithm of jointing the color features and texture features is proposed.Firstly,transformed the color space to HSV space.Secondly,utilized simplified local intensity order pattern and completed local binary pattern(CLBP)to extract the color features in the H channel and the texture features in both S channel(Saturation)and V channel(Value)respectively.Thirdly,concatenated these features as the descriptors of a color texture image.At last,adopting SVM for classification.Based on experiments applying to the popular image datasets,it can be conclude that the algorithm possesses the advantage of lower dimension and better computational efficiency and identification precision than other classical algorithms.On KTH-TIPS2 b and CUReT databases,the classification accuracy of the proposed algorithm is 99.4% and 99.6% respectively.Experiment results prove that the proposed algorithm possesses rotation invariance and illumination robustness when it is used classification of color texture images.Aiming at the disadvantage of the Spatio-Temporal Context(STC)in tracking,which is not strong in anti-occlusion ability and prone to drift,the improved algorithm was proposed.Firstly,estimated the type of occlusion of the target,and then applied different model update strategies for different occlusion types and the weighted confident map of the block.In this way,the occlusion information will not update to the target model when the target is occluded.Using the improved algorithm,the experiments are conducted on the Visual Tracker Benchmark.The results show that the improved algorithm has a great improvement compared with the original STC algorithm in anti-blocking performance and is also robust to illumination changes.Besides,the improved algorithm can also meet the real-time requirement.