Dimensionality Reduction Technique on SIFT Feature Vector for Content Based Image Retrival

Abstract

Scale-Invariant Feature Transform (SIFT) descriptors plays very important role for depicting and matching the digital images between various views. There are various practical issues like keypoint localization and image retrieval that uses the SIFT descriptors for matching the image content among different views. The problem associated with these descriptors of an image is that calculation and matching of SIFT features descriptors are very cumbersome and slow process.For removing this problem the proposed method presents a technique that reduces the complexity, size and the time for matching of SIFT descriptors used in robot localization and indoor image retrieval.The proposed method reduces the number of SIFT descriptors and the complexity of every SIFT descriptor for determining an image. Our outcomes demonstrate that there is a negligible loss of exactness in feature retrieval while accomplishing a critical decrease in picture descriptor estimate and coordinating time. Proposed technique diminishes the descriptor size (number of descriptors) and furthermore speed up the searching of an image from extensive dataset. Decreasing the descriptor estimate also reduces the storage space required for storing the image descriptor.