The accuracy for any CBIR structure depends on the accurate recovery of relevant images from a larger database. This paper creates a new derivation for existing texton methods to enhance precision in CBIR. No method on textons defined any sub textons primarily based on the magnitude relation between pixels that are part of textons and the rest (not part of the textons). This paper proposed a new variation to texton called magnitude-based twin texton co-occurrence matrix (MTTCM), which explores micro structures of the 2 x 2, and integrates the color, texture, structural features with edge information. The MTTCM is evaluated on natural texture datasets, and the findings show that it outperforms current representative image feature descriptors.