A Low-illumination Image Enhancement Algorithm Based on Retinex to Improve SIFT Matching in Similar Scenes

A low-light enhancement algorithm based Retinex model is proposed, which has the effect of improving SIFT matching on low-light image. Firstly, according to Retinex model, Illumination map is obtained by Gaussian filtering. Secondly, the original image is subtracted from the illumination map, and the result of this processing is corrected using gamma correction. Finally, add the original image after Histogram Equalized to the image obtained by the second step. The results of Mikolajczy database show, the number of SIFT feature points were 2709, 2294, 1969, 1719, 1532, and 1240. After image processed by this algorithm, the SIFT feature points extracted are 6126, 5656, 5349, 4926, 4574 and 4202. The matching accuracy was also improved from 0.92, 0.91, 0.87, 0.86, 0.77 to 0.94, 0.93, 0.92, 0.89, and 0.86.