K-means and Morphological Approach on Image Segmentation for Fish Detection

Object detection using image segmentation by divide the image into several parts, such as the background and the object. However, object detection is difficult if the object and background have similar colors. This research performs image segmentation using K-means clustering to detect fish based on color and grayscale images. Before the segmentation, the acquired image needs to be preprocessed to increase image quality the image preprocessing used methods like cropping, resizing, and image enhancement. 

The preprocessing results become input in the K-means segmentation. This method used K-means clustering, dividing (cluster) images into several parts by Euclidean distance. In the last step, the K-means segmentation is processed by morphology (opening and closing methods) to get the contours of the black-white (BW) fish image from color and grayscale. 

Based on the experiments, as many as 100 fish images can be detected using the best k value is 8, and the accuracy is 84.48% to detect fish. Besides that, the structural similarity index measurement (SSIM) calculation returns values close to 1. This result means that the information is almost the same as the fish image. This segmentation can detect fish images with clearly segmented fish images.

 Keywords— fish detection, histogram equalization, image processing, image segmentation, k-means clustering, morphology.