A Unified Solution for CNN Template Construction with RWC Algorithm

In this paper, we present a solution for constructing the weights of a Cellular Neural Network (CNN) through the use of a Random Weight Change (RWC) algorithm. The process begins by creating a target image for each input image through image processing techniques such as sketching. The original weights are then adjusted by adding a randomly generated small value vector, and the results are tested with the input-target image pair. If the learning error decreases, the weight is used in the next iteration and updated with the same random vector. If not, a new random vector is generated. The proposed weight learning method has several advantages, including its simplicity and straightforward learning algorithm, as well as its compatibility with all types of CNNs and input images without requiring modifications to the original CNN structure. The success of this method is demonstrated through the successful learning of templates for various image-processing tasks using different CNN structures.