A multiscale retinex with low computational cost

In this paper, we propose a multi-scale retinex (MSR) in the wavelet transformation domain. Retinex method mainly of two steps: estimation and normalization of illumination. Illumination has only low-frequency components. The estimation of illumination can be performed by using the coarse component in the wavelet expansion of input image. Therefore, high-frequency components in the wavelet expansion are preserved. The proposed Retinex can preserve the high-frequency components of the input image, and get a better enhanced image compared to the output of standard MSR. The output of standard MSR is a weighted sum of several different SSR outputs. Thus, the MSR algorithm has still heavy processing to deal such as motion pictures because of Gaussian filter with wide surround space is used. On the other hand, in our method, Gaussian filtering is applied to half or quarter the size of original image. Furthermore, the surround space of Gaussian filter can be set small. Thus, the computational cost is only 1% of the standard MSR.