Multi-level Thresholding Using Adaptive Gravitational Search Algorithm and Fuzzy Entropy

Conventional multilevel thresholding methods are computationally expensive when applied to color images since they exhaustively search the optimal thresholds by optimizing the objective functions. To address this problem, this paper presents an adaptive gravitational search algorithm (AGSA) based multi-level thresholding for color image. In AGSA, a dynamic neighborhood learning strategy which incorporates the local and global neighborhood topologies is introduced to achieve adaptive balance of exploration and exploitation. Moreover, a sinusoidal chaotic based gravitational constants adjusting operator is embedded to further promote the performance of AGSA. When extending AGSA to solve the multi-level thresholding problem, the fuzzy entropy is adopted as the objective function. Experiments were conducted on two color images to investigate the efficiency of the proposed method. The obtained results are compared with that of the particle swarm optimization (PSO) and gbest-guided GSA (GGSA). The experimental results are validated qualitatively and quantitatively by evaluating the mean of the objective function values and the total CPU time required for the execution of each optimization algorithm. Comparison results showed that the AGSA produced superior or comparative segmentation accuracy in almost all of the tested images and the algorithm largely reduce the computational efficiency of GSA.