Video Image Retrieval Based on Improved SIFT Features

Abstract

This paper presents an algorithm based on SIFT features. It calculates key points in the image and extracts the feature of the image by calculating the key points' orientation and modulus of the gradient. The similarity between two images is computed using Euclidean distance. The experiment shows that the feature is invariant to image scaling translation, rotation, and partly invariant to illumination changes and has a certain affine invariance. It is better than the color feature in the video image retrieval.