Effective feature detection and object recognition based on local invariant feature detector

Abstract

Object recognition using local feature is a research area with rapid progress. The recognition of an object has regained great attention in computer vision and artificial intelligence, especially for different imaging conditions. Object recognition empowers major real-world applications such as object tracking, video surveillance, robotic vision and agricultural application etc. Many object recognition techniques have been developed by researchers during the past few decades. Object recognition has limitations when it comes to varied scale, viewpoint, illumination and JPEG compression. The local feature extraction method for recognizing an object is more effective than global feature extraction method. Local feature extraction method effectively handles changes in scale, rotation, illumination and JPEG compression. Great strides have been made in this field of local feature extraction during the last few years, there are still many avenues that need to be pursued and remain challenging for computer vision researchers. This thesis focuses on object recognition using invariant feature extraction method and analyses various feature detector and descriptor along with the proposed methods. The main objective of this research work is to propose invariant feature detector for object recognition system. The most commonly used feature detector and descriptor is Scale Invariant Feature Transform (SIFT). The use of Gaussian filter in the SIFT algorithm fails to match feature points on the edge. The extraction of invariant feature point detector using Trilateral filter with a Harris Corner feature Detector (THCD) preserves the high frequency content of the image. This method enables the extraction of invariant feature point detection for various image transformation such as scale, rotation, illumination, viewpoint and JPEG compression. To improve the performance of feature point detectors Harris corner point is added. THCD descriptor is formed and evaluated based on recall vs 1-precision curve to prove the robustness of the