Fish Detection Using Morphological Approach Based-on K-Means Segmentation

TL;DR

This paper presents a morphological approach using enhanced k-means segmentation for effective fish detection in images, addressing challenges posed by complex backgrounds and similar colors.

Contribution

The study introduces a combined preprocessing, k-means clustering, and morphological operations method for improved fish detection in images.

Findings

High SSIM values indicate minimal information loss.

Clear fish contours achieved through morphological processing.

Effective segmentation in complex backgrounds.

Abstract

Image segmentation is a concept that is often used for object detection. This detection has difficulty detecting objects with backgrounds that have many colors and even have a color similar to the object being detected. This study aims to detect fish using segmentation, namely segmenting fish images using k-means clustering. The segmentation process is processed by improving the image first. The initial process is preprocessing to improve the image. Preprocessing is done twice, before segmentation using k-means and after. Preprocessing stage 1 using resize and reshape. Whereas after k-means is the contrast-limited adaptive histogram equalization. Preprocessing results are segmented using k-means clustering. The K-means concept classifies images using segments between the object and the background (using k = 8). The final step is the morphological process with open and close operations to obtain fish contours using black and white images based on grayscale images from color images. Based on the experimental results, the process can run well, with the ssim value close to 1, which means that image information does not change. Processed objects provide a clear picture of fish objects so that this k-means segmentation can help detect fish objects.