The invention claimed is:1. A computer-implemented method for generating a color representation, the method comprising:a receiving an input image;determining a color representation for each region of the input image using a trained artificial neural network, the trained artificial neural network comprises a U-net architecture comprising a first neural network and a second neural network, the trained artificial neural network taking as input a histogram of the respective region and outputting the respective color distribution, second neural network trained to determine alpha masks to determine regions of the image associated with each color representation, each color representation having one of three base colors at each respective vertex of a triangular face, the triangular face having a color distribution therein, the color distribution discretized into discrete portions, the amount of discretization based on a patchwork parameter, each discrete portion having an interpolated color determined to be a combination of the base colors at respective coordinates of such discrete portion, the first neural network trained to encode and decode the histogram for each alpha mask to determine the base colors of the color distribution, discretization of the color distribution, and the patchwork parameter; andoutputting the color representations.2. The method of claim 1, wherein the first neural network is trained to minimize an error metric between colors of the respective region of the input image and the three base colors.3. The method of claim 2, wherein the colors of the respective region of the input image are represented by a deformed planar patch of Red-Green-Blue space.4. The method of claim 1, wherein the alpha mask comprises a grayscale image, and the sum of all alpha masks for each pixel is 1.5. The method of claim 1, wherein the combination of the base colors for each discrete portion comprises a linear combination of the base colors.6. The method of claim 1, wherein the first neural network further determines a wind parameter, and wherein blending behavior of the color distribution is determined based on the wind parameter acting at a focal point.7. The method of claim 1, wherein the blending behavior uses a cubic Bezier triangle.8. The method of claim 1, wherein a softmax function is applied across channels at every pixel of the alpha masks.9. A system for generating a color representation, the system comprising one or more processors and a data storage, the one or more processors configured to execute:a mapping module to use a trained artificial neural network to determine alpha masks each representing a region of a received input image, each alpha mask associated with a color representation, the trained artificial neural network comprises a U-net architecture, the U-net architecture comprises a first neural network and a second neural network, the second neural network trained to determine the alpha masks for each region; anda representation module to determine the parameters of the color representation associated with each region of the input image using the trained artificial neural network, the trained artificial neural network taking as input a histogram of the respective region and outputting the respective color distribution, each color representation having one of three base colors at each respective vertex of a triangular face, the triangular face having a color distribution therein, the color distribution discretized into discrete portions, the amount of discretization based on a patchwork parameter, each discrete portion having an interpolated color determined to be a combination of the base colors at respective coordinates of such discrete portion, and outputting the color representations, the first neural network trained to encode and decode the histogram for each alpha mask to determine the base colors of the color distribution, discretization of the color distribution, and the patchwork parameter.10. The system of claim 9, wherein the first neural network is trained to minimize an error metric between colors of the respective region of the input image and the three base colors.11. The system of claim 10, wherein the colors of the respective region of the input image are represented by a deformed planar patch of Red-Green-Blue space.12. The system of claim 9, wherein the alpha mask comprises a grayscale image, and the sum of all alpha masks for each pixel is 1.13. The system of claim 9, wherein the combination of the base colors for each discrete portion comprises a linear combination of the base colors.14. The method of claim 9, wherein the first neural network further determines a wind parameter, and wherein blending behavior of the color distribution is determined based on the wind parameter acting at a focal point.15. The system of claim 9, wherein the blending behavior uses a cubic Bezier triangle.16. The system of claim 9, wherein a softmax function is applied across channels at every pixel of the alpha masks.
