Performance Analysis of SIFT/SURF Algorithms in Neural Networks for Optimized Feature Detection

Abstract

This paper is an experiment on the implementation of scale-invariant feature transform (SIFT) and speeded up robust features (SURF) algorithms into multi-dimensional neural networks. We are attempting to perform a comparative performance evaluation by using different scale factors of the SIFT algorithm in multi-layered neural networks. This method will help us to understand the best way of implementing the above algorithms in neural networks and from a given sample, extracting distinctive invariant features and finding points of interests. Hence performing a large data set computation would be made much easier because of the neural network implementation. The conventional method of performing SIFT has computational limitations and we aim to achieve best possible way of performing the feature detection when using SIFT and neural network combined, hence transcending computational limitations that SIFT previously had. This approach to recognition of features can robustly find results much faster on bigger dataset and at the same time have the benefits of SIFT algorithm.