A Novel Framework for Optimal RGB to Grayscale Image Conversion

Nowadays most images are shot as color images. Yet in situations such as printing or pattern recognition they also need to be converted to grayscale images. The most important problem in this conversion process is to preserve the image contrast. In this paper we make two contributions. In the simple yet popular line projection approach, which is also adopted in Matlab, we propose an entropy-based optimization framework to choose the optimal line direction so that all the pixel color vectors in an image have the most spread-out projections, thus increasing the grayscale image contrast. Secondly, we make use of histogram specification on all the projection points to further increase the image contrast. Experimental results show that the proposed framework produces enhanced results compared to typical other methods.