A method of recursive target extraction based on multi-level features

Abstract

Due to the complexity and asymmetrical illumination, some images are difficult to be effectively segmented by some routine method. In this paper, an algorithm based on multi-level features is designed and proposed, and which can be used for target extraction from the images with more noises, interference, uneven illumination and changeable scene. The algorithm first transfers the original image into a gray one. And then features of every level the target are extracted inheritably from the high or low level feature message. Furthermore, it also can track back to the original image or the features of low level, and the extraction goes on by recursion. So the target can be separated from the background. The algorithm experiment results indicates the target can be correctly extracted with high-efficiency and great precision, and with different sizes of the target and SNR also.