Infrared image enhancement algorithm based on multiscale retinex with adaptive surround space constant

Abstract

Retinex is an image enhancement algorithm based on color consistency theory which is originally used for color image enhancement with a trade-off between dynamic range compression and tonal rendition, and these two effects are combined with a weighted sum of several Single Scale Retinex (SSR) outputs, namely Multi Scale Retinex (MSR) algorithm. For infrared image, it is known that the resulting effects of dynamic range compression and detail enhancement are governed by the Gaussian surround space constant. Traditional MSR algorithm using fixed surround space constants and can not automatically adjust each surround space constant for different images. This means, MSR algorithm can not improve the image details and overall visual effects at the same time using local information of infrared images. To overcome this problem, an MSR based image enhancement algorithm with adaptive surround space constant is proposed. Firstly the image standard deviation is calculated to scale overall dynamic range, then the image is divided into several sub-images, standard deviation and mean value of gradient for each sub-image are calculated to measure the local complexity. The surround space constant in MSR algorithm is determined by both overall dynamic range and local complexity that is acquired above to achieve an optimal enhancement output. Then intensity stretching is used to produce a graceful visual effect. Experiment proved that the algorithm proposed in this paper improves the quality of infrared images effectively, and has better adaptability than traditional MSR algorithm.