Detection and recognition method of imaging objects and perception computation based on vision bionics

Object detection and identification are core issues in image analysis and understanding. This paper constructed system framework and computation model for vision detection and object recognition based on vision bionics. Inspired by the human visual information acquisition, variable spatial resolution mechanism, and sparseness of processing, we constructed large filed subsystem ( LF ) and small field subsystem ( SF ) to obtain primary vision vectors of multi-resolution, multi-scale, and different fine scales, respectively. The paper proposed an object detection and identification method, which first using LF subsystem to perceive the statistical characteristics of whole scene in wavelet domain guided by vision attention mechanism. And then the SF subsystem gazed at the object and extracted feature vectors of fine scales, and combined the features into a saliency map. Finally, non-uniform sampling, multi-scale analysis, and winner-take-all mechanism are used to generate competition between targets to realize correct classification. The experimental results demonstrate that the statistical method reduces redundancy, and can detect the interesting region quickly and exactly. Furthermore, the attention mechanism based object recognition method reaches an average accuracy rate of 94.40%.