Colour correction of dermatoscopic images based on Shades of Gray and the Adaptive Multiscale Retinex Method

Abstract

Melanoma is the most common and deadly malignant skin cancer. A method based on colour correction and brightness adjustment was proposed to solve the colour deviation problem of dermatoscopic images in melanoma recognition. Firstly, the Minkowski norm is used to correct the colour deviation in the red channel of the dermatoscopic image. Secondly, the image’s saturation is adjusted by the linear stretching method. Then, the convolution operation of the guided filter was used to estimate the incident component of the low illumination image, and the reflection component formula of the guided filter based on the multi-scale Retinex method was constructed by adjusting the size of the filtering window and the value of smoothing parameters to enhance the brightness of the image. Finally, the Gamma correction algorithm adjusts the brightness image appropriately. The experimental results show that the proposed algorithm can solve the problem of abnormal colour deviation of dermatoscopic images and improve the recognition rate of melanoma by 4-6 percentage points through colour correction.