Using Whale Optimization Algorithm and Haze Level Information in a Model-Based Image Dehazing Algorithm.

Area of Science:

• Computer Vision
• Image Restoration
• Artificial Intelligence

Background:

• Single image dehazing is a challenging problem in computer vision.
• Model-based methods, like dark channel prior (DCP), are effective but require parameter tuning.
• Previous work introduced an improved DCP (IDCP) using heuristic scaling factors.

Purpose of the Study:

• To develop an optimized dark channel prior (OIDCP) algorithm for enhanced single image dehazing.
• To automatically determine optimal model parameters for dehazing using optimization algorithms and haze level information.
• To improve the robustness and performance of model-based dehazing methods.

Main Methods:

• Utilized the whale optimization algorithm (WOA) to find optimal scaling factors for the IDCP method.
• Developed a hazy image discriminator to exclude unsuitable ground truth images.
• Implemented hazy image clustering to group images by haze level and derive average optimal scaling factors for the OIDCP.

Main Results:

• The proposed OIDCP algorithm achieved superior performance on RESIDE, O-HAZE, and KeDeMa datasets compared to DCP, IDCP, and other recent methods.
• On the RESIDE dataset, OIDCP achieved a PSNR of 26.23 dB, outperforming IDCP by 0.81 dB and DCP by 8.03 dB.
• The OIDCP demonstrated stable visual quality and consistent performance across different haze levels.

Conclusions:

• The OIDCP algorithm effectively optimizes model parameters for single image dehazing, significantly improving restoration quality.
• The integration of WOA and haze level analysis provides a robust approach for parameter optimization in model-based dehazing.
• The findings suggest that the OIDCP method can benefit the broader field of model-based dehazing algorithms.