Multilevel thresholding for crop image segmentation based on recursive minimum cross entropy using a swarm-based technique

Abstract

The segmentation of crop images is the most sought-after technology in the field of computer vision and pattern recognition. The multilevel thresholding (MLT) of crop images is a tedious job due to the complex background and uneven distribution of intensity levels. The recursive minimum cross entropy (R-MCE) approach is a notable technique for MLT. Therefore, in this paper, R-MCE has been used with an efficient cuckoo search algorithm (CSA), based on Lévy flight for determining the best thresholds. This technique reduces the complexity of computation and increases the accuracy, when it is used for MLT. The dexterity of the proposed technique has been examined over 20 crop images with different illumination and complex backgrounds. Results of the proposed method have been expressed in terms of feature similarity index (FSIM), structural similarity index (SSIM), mean square error (MSE), peak signal-to-noise ratio (PSNR), and CPU time. To investigate the efficiency of proposed technique, the segmented results have been compared with well-known thresholding methods, such as wind-driven optimization (WDO), bacterial foraging optimization (BFO), beta differential evolution (BDE), and artificial bee colony (ABC). Simulation results show that the MLT of the proposed approach can accurately and efficiently segment the complex background crop images.