A method, computer program, and computer system are provided for encoding video data. Luma and chroma samples are identified from image data corresponding to a video frame. In particular, at least one chroma sample may be predicted from at least one luma sample based on a DC contribution from a chroma component of the image data and an AC contribution from a luma component of the image data. Also, a plurality of linear models may be determined from the identified luma and chroma samples. The luma and chroma samples may be classified into a number of categories corresponding to a number of linear models, where each category has at least one corresponding linear model parameters signaled to the bitstream; the chroma sample prediction may be further based on such parameters.