Image Enhancement Algorithm of Vehicle Recognition under High Light Based on Deep Learning

An adaptive image enhancement algorithm under high light conditions is proposed to solve the complex light problem of traffic checkpoint surveillance cameras. This algorithm mainly includes three modules: rapid image classification module, image enhancement module and image quality evaluation module. The rapid image classification module divides the high light image into daytime high-light type and night high-light type on the basis of SqueezeNet network of deep learning, reducing the computation burden while ensuring the classification accuracy. The image enhancement module makes use of the integrated image enhancement algorithm for secondary enhancement of the classified images by adopting the separate handling thought. The image quality evaluation module adopts objective evaluation indexes to measure and designs a weighted comprehensive evaluation index, which is measured from three aspects, namely, structural similarity, normalized mutual information, and normalized mean square error. The experimental results indicate that the proposed algorithm can still obtain a relatively accurate recognition rate for high-light images under different conditions, improving the subsequent recognition of vehicle faces and vehicles.