FPGA-based favourite skin colour restoration using improved histogram equalization with variable enhancement degree and ensemble extreme learning machine

Abstract

This paper presents skin color enhancement based on favorite skin color to agree with user-defined favorite skin color using improved histogram equalization with variable enhancement degree (IHEwVED) and machine learning methods. The skin color to be adjusted in the input image is shifted to favorite skin color by using novel control parameters of the proposed IHEwVED method. Three different novel display device-dependent color image processing methods are introduced based on hsv and yiq color space to obtain the desired enhanced output images. A reduced convolutional neural network and the novel ensemble extreme learning machine (EELM) architectures are developed and implemented in a field-programmable gate array to test, synthesize, and validate the recognition capability of the user-defined favorite skin color. The less computational complex proposed IHEwVED-EELM method recognizes 45 to 50 favorite skin color per second of test images by consuming 0.035 second training time with training root mean square error (RMSE) of 0.0048 and testing RMSE of 0.01208. Finally, a stand-alone favorite skin color restoration system is developed using the high-speed video processor NI-PXI-1031 based on the IHEwVED-EELM method in the Python-OpenCV environment. The laboratory experimental performances ascertain the real-time ability of the proposed favorite skin color restoration method.