An Approach for Target Detection and Extraction Based on Biological Vision

Inspired by the mechanism of multi-scale image fusion of insect compound eye, this paper proposed a target detection and extraction method based on insect compound eye and human visual attention mechanism. The main feature of this method is that multi-scale visual attention mechanism is designed for improving the detection accuracy of interested target, meanwhile image is pre-segment based on it, then the target is extracted based on the LEGION (Locally Excitatory Globally Inhibitory Oscillator Networks) selection network model. For multi-scale visual attention mechanism, on intensity channel, this approach adopts multi-scale analysis to get the intensity saliency map on both global-level and local-level; on orientation channel, according to the mechanism of the large vision field of compound eye, directional information is added to improve the orientation sensitivity; on color channel, according to the characteristic of color sensitivity of human vision system, HSI color space is used instead of RGB to enhance color features. Compared with the experiment results of Itti visual attention model, the proposed approach can locate the interested targets in clutter scene more accurately, and complete fast extraction of the target without increasing of computation and time-consuming.