Recovering intrinsic images: An evolutionary technique for entropy minimization

Shadow removal from color images is considered as a challenging task during last decades. Several approaches have been introduced to address and improve this task. A major breakthrough in this area is projecting the correct direction that minimizes the entropy to get rid of the lighting effect. In this work, we reduce the computational time of entropy minimization up to 52% and decrease the number of integrations by a factor of three using genetic algorithm. The first population is generated by considering the distribution of training images. Then, crossover and mutation are applied and after few generations, the algorithm can reach to the minimum entropy. In the second contribution, another system is proposed based on genetic algorithm that pave the way for finding a real number, instead of integer number, for entropy minimization. Thanks to the new method, the result shows that a real number for the angle can be accurately found on the reasonable time.