Improved Image Local Descriptor

Abstract

Detecting the similarity in term of feature point in different images is the fundamental step in computer vision application. Similar features matching make the task of object tracking, robot localization, pose estimation, action detection much easier. Several feature detection and description techniques have been developed in the literature. Depending on the computation time and recognition rate these techniques are used in different application. Feature descriptor is used to describe the image interest points in such a way that the description remains robust against geometric and photometric transformations of image. In this work, the performance of the image local descriptor difference of polynomial is further improved against low frequency signal variations (that are 3D view angle variations, rotation, illumination changes and image blur). The orientation detection using dominant orientation shift algorithm improves the matching accuracy. The patches illumination variations are equalized using weighted threshold histogram equalization scheme. The patch is subjected to Guided image filter to smooth the textures while preserve the edges in the extracted interest region. The use of correlationcoefficient instead of Euclidian distance for descriptor matching has improved the matching accuracy. Experimental results show that our proposed changes have improvedthe performance of the Difference of polynomial feature descriptor and had made the descriptor to categorize the feature with more distinctiveness and robustness against geometric and photometric transformations of image.