Method and system for color representation generation
摘要
說明書
In a particular case of the system, the set of parameters further comprising a focal point parameter and a wind parameter.
In another case, the color distribution comprises a histogram of bins in Red-Green-Blue (RGB) color space provided as input to the trained neural network.
In yet another case, the bin values are normalized.
In yet another case, input to the trained neural network comprises error metrics on pixel colors in the input image.
In yet another case, the error metrics comprise a greedy Red-Green-Blue (RGB) reconstruction loss and a fraction of colors in the color distribution that have a relevance to the color representation above a predetermined threshold.
In yet another case, the mapping module further determines the relevance comprising performing Kullback-Leibler divergence.
In yet another case, determining the divergence further comprises determining histograms for patches of the input image; taking a maximum for each patch; and normalizing the maximums.
In yet another case, the mapping module further maps each portion of the color representation with a color that is closest to a pixel or group of pixels in the input image; and wherein the representation module further: receives a modification to any one of the set of parameters; regenerates the color representation with the modified parameter; and recolors each pixel or group of pixels in the input image to the color value of the mapped portion of the color representation.
