Saliency detection based on hybrid artificial bee colony and firefly optimization

In recent years, saliency detection in images has become a major image processing field. The human eye's focusing function is emulated in salience detection, which aims to discover the first focused area in images. The method of detecting salient areas can be utilized instead of the models established to handle problems like object recognition, image segmentation, and video surveillance. As a result, salience detection is a significant aspect of image processing. The use of optimization techniques to solve image processing difficulties is a popular research topic. In this study, a hybrid approach (hybrid artificial bee colony firefly algorithm – HABCFA) was developed by combining artificial bee colony (ABC) and firefly (FA) optimization methods to answer the problem of salience detection in images. Bottom-up and top-down approaches are two types of algorithms created to find the most important area or object in an image. Top-down methods require learning-based systems, whereas bottom-up methods rely on data received from images. This study's approach is classified as bottom-up because it makes use of the image's color information. During the preprocessing stage, the SLIC approach, which is a super-pixel-based classification method, was utilized to extract salient regions more precisely and obtain a more ideal background value for the salience detection process. Following the preprocessing stage, the HABCFA method is used to optimize the background value obtained from the image's edge regions, resulting in the best acceptable background value for the image. The ideal background value is subtracted from the current image to produce a saliency map. In the experiments, using the popular MSRA-1000, ECSSD, ICOSEG, and DUTOMRON datasets, the HABCFA salience detection approach was compared to 11 state-of-the-art methods in the literature. When the results of HABCFA without any additional masking or training steps were compared to the results of other approaches, it was discovered that HABCFA was effective. In addition, in terms of convergence rate and running times, the ABC, FA, and HABCFA methods were compared to four known comparison problems, with HABCFA outperforming the other two optimization methods.