We propose a new method to enhance the cloth stain, such as the weak coffee stain in image. Our method is composed of two steps. At the first step, we obtain the principal component images by making the RGB components of each pixel to be uncorrelated. At this step, the texture structure is still left in the uncorrelated images, so it is difficult to recognize the stain. Then, at the second step, we reduce the texture structure by using Principal Component Analysis. We divide the uncorrelated images into the blocks, and those blocks are approximated by several principal eigen vectors which represent the texture structure. The approximation image represents the average texture structure without the cloth stain. So, the cloth stain is recognized without endeavor in the residual image which is obtained by subtracting the approximation image from the principal component image. As our method is robust for the number of observation data, we need only one color image whose size is not so big and whose resolution is not so high. The results of our experiment under different conditions show that the proposed method is effective for the image enhancement of the small weak cloth stain.
View full abstract