Traffic sign recognition method with biologically inspired transform

Rotation, scaling, translation and noise (RSTN) invariant feature extraction is a valuable technique for practical traffic sign recognition. However, it is difficult to deal with rotated, scaled, translated and noised traffic sign images for conventional methods. Neurons at higher levels exhibit an increasing degree of invariance to image variation. Moreover, visual cortex response to some directional light bars strongly. In terms of these mechanisms, a biologically inspired transform (BIT) approach for traffic sign recognition is proposed in this paper. Like human visual cortex, BIT includes two stages. In the first stage, we build an orientation edge detector to highlight the edges of different directions. The orientation edge detector is primarily composed of a phase congruency based edge detector and a bipolar filter. Then, a local spatial frequency detector produce a response pixel, which also converts rotation or scaling of orientation edge into a horizontal or vertical shifted map. In the second stage, the orientation edge detector and local spatial frequency detector are performed again, which converts shifted map into an invariant pixel in the final map. We have performed some experiments to illustrate the effectiveness of the extracted invariant traffic sign features. The results provide supports for that our approach can achieve better recognition performance under rotation, scaling, translation, noise interference and even illumination diversification.