A Proposed ICA Algorithm

The next process is to do iteration to meet convergence. The initially assumed weights are sent to the normalization unit after updating. Convergence is confirmed by the convergence confirmation unit. When the convergence threshold is met or the maximum iteration is reached, the iterative process ends and the data is sent to the separation matrix estimator to estimate the source signal. Otherwise, the adaptive optimization unit checks whether the fitness parameter has a positive or negative value.

This paper proposed a cooperative algorithm based on AFSA and k average. This algorithm is used to perform multilevel thresholding based on histogram. In the proposed algorithm, artificial fish (AF) first performs the optimization process in AFSA. After the cluster converges, the cluster center obtained by AF is used as the initial cluster center of the k-means algorithm. After transferring the output of AFSA to k-means, AF reinitializes the cluster and executes it again. In fact, in the proposed algorithm, AFSA is used for global search and k mean is used for local search. The proposed algorithm, like the other four algorithms, is used for image segmentation of two known images, Lenna and Barbara. Efficiency comparison showed that the algorithm has appropriate and acceptable efficiency

This section describes the proposed algorithm. In the proposed algorithm, there is AFS group of AFSA. These AF groups are initialized randomly in the problem space. Each AF consists of K cluster center positions in the one-dimensional image histogram space. Therefore, the search space of AFSA for the K cluster center has K components. The fitness function AFSA has to minimize is shown in equation (3). The histogram is clustered according to equation (3) based on the color distribution between the pixels of a given image. The image is divided into K clusters (Ci) by K - 1 thresholds according to color attributes. In Equation (3), the frequency (f j) of the pixel having the color value X j on the given image is added to the distance (Z i) between the color X j on the image histogram and the center of the cluster to which it belongs, Is multiplied. Calculate this value for all color values ​​relative to the center of the cluster you are belonging to.

This paper proposed a cooperative algorithm based on k-means for artificial fish school algorithm and image segmentation by multilevel thresholding. In the proposed algorithm, AFSA performs global search, and k average is responsible for local search. The proposed algorithm's process makes it possible to prevent robustness and the ability to be trapped in local optimization. We divided the two well-known images using the proposed algorithm and the other four algorithms and compared the results obtained with each other. Experimental results show that the segmented image quality of this algorithm is much better than the other four test algorithms.