A multi-scale retinex algorithm for image enhancement

Rapidly and robust image enhancement transformation is important for numerous applications in intelligent video surveillance, which can improve the quality of scene image and extract better features in complex environment where the images undergo large lighting changes. In this paper, a multi-scale retinex algorithm is presented for image enhancement, which has two contributions. First, the initial approximation image is computed by both each pixel value and maximum value of original image. Second, discrete wavelet transformation is used to decrease computation complexity. Experimental tests on numerous scene images show that proposed algorithm has better performance for edge detecting from surveillance images and lane images.