Image Feature Extraction and Selection
Features are the description of the image contents which could be corner, blob or edge. Scale-Invariant Feature Transform (SIFT) extraction and description patent algorithm used widely in computer vision, it is fragmented to four main stages. This paper introduces image feature extraction using SIFT and chooses the most descriptive features among them by blurring image using Gaussian function and implementing Otsu segmentation algorithm on image, then applying Scale-Invariant Feature Transform feature extraction algorithm on segmented portions. On the other hand the SIFT feature extraction algorithm preceded by gray image normalization and binary thresholding as another preprocessing step. SIFT is a strong algorithm and gives more accurate results but when system require increasing speed, it is better to select distinctive features and use them in description process. The experimental results show clearly reduction of features extracted using SIFT algorithm on segmented parts and the algorithm of feature extraction from normalized binary image gives better results for feature localization as shown in experimental images.