A Two-Level Approach to Color Space-Based Image Segmentation Using Genetic Algorithm and Feed-Forward Neural Network

Abstract

This work proposed an approach for the segmentation of high-resolution images such as satellite imagery stand on the supportive method of GA and FFNN. During this two-layer technique, the GA applies for the selection of the best individual (pixels in image). Based on the outcome generated by this process, feed-forward neural network is trained to carry out the detection and segmentation. Neural network is trained using an approach called Levenberg–Marquardt algorithm. To improve the quality of the segmentation process, the original test image is transformed into various color spaces and then segmentation is applied. In this work, bivariate image value actions are utilized to validate the excellence of the output (segmented) image based on the assessment of subsequent image pixels between input and output (segmented) images. The investigational outcome illustrates the effectiveness of the two-layer process of GA and FFNN in support of the segmentation of high-resolution images. The experimental analysis based on the image quality measures exposed the crucial role of color space for the proposed work.