K_means method is commonly used for color image clustering, and it can be used also for color image features creation, these features can be used later on as an image signature which refers to the image in any application, such image recognition or image retrieval system. Using clusters or within_cluster sums created by K_means method is not correct because the values of these sets change from time (run) to time. So a modification was added to this method, by proposing a new method. The proposed method will be implemented, the clusters for each image will be created, also the within_cluster sums will be created and it will be shown that each of these data sets can be used as an image features to create an image signature.