Feature extraction method based on biological vision

Abstract

The invention relates to a feature extraction method based on biological vision. The method comprises the steps that multi-direction multi-scale Gabor filtering is carried out on an original input image to obtain a set of responses of the original input image, and the responses are used for simulating locality and direction sensitive features of a simple cell receptive field of a visual cortex; the maximum response value and the direction corresponding to the maximum response value of each position of the responses are obtained, and are used for simulating invariance extraction features of a complex cell receptive field of the visual cortex; the responses are divided into a plurality of square blocks equal in size, histogram statistic is carried out in the direction with the square blocks as the unit, and statistics is carried out on each position in the maximum response value direction of each position to reflect overall space distribution performance of the features; the accumulation and standardization operation are carried out with the square blocks as the unit to obtain a set of feature vectors with selectivity and invariance. The method simulates the features of the simple cell receptive field and the complex cell receptive field of the visual cortex, low-dimension and effective features are obtained for images of complex scenes, and the method is used for the field of object identification, detection and tracking.

Description

technical field [0001] The invention belongs to the cross technical field of computer vision and biological vision, and relates to an image feature extraction method, in particular to a feature extraction method using the receptive field characteristics of simple cells and complex cells in the visual cortex. Background technique [0002] Images in complex scenes have the basic characteristics of high dimensionality and redundant information. How to effectively reduce its dimensionality, remove redundant information, and obtain low-dimensional and effective features is a decisive step. It can be said that the quality of feature extraction determines the upper limit of the ability to solve problems, and the subsequent models and methods based on feature extraction are only for better or more effective approach to this upper limit. [0003] The traditional feature extraction method is actually manually designing features, often targeting a specific field. Although there are man...

AI Technical Summary

Problems solved by technology

The mode of simulating the receptive fields of simple cells and complex cells in the visual cortex described in this patent lacks biological visual evidence (see "T. Serre, L. Wolf, S. Bileschi, M. Riesenhuber, T. Poggio, Robust Object Recognition with Cortex- Like Mechanisms, IEEE Transactions on PAMI29(3)(2007)411-426."), its robustness is yet to be verified

In addition, the patent abandons the multi-scale analysis of the image for the sake of efficiency, which will inevitably lead to the lack of scale invariance of the extracted feature vector

Images