Optimized Defogging Algorithm for Dark Channel Based on Multi-Scale Convolutional Neural Network

To address the issue of significant discrepancies between the estimated transmission in traditional single-image dehazing algorithms and the actual scenario, a multi-scale convolutional neural network-based dehazing algorithm optimized by dark channel prior is proposed. The algorithm utilizes guided filtering, which preserves edge smoothness, to filter the feature transmission map extracted by the convolutional neural network, making the processed transmission closer to the actual transmission of the input hazy image. Subsequently, the dark channel prior theory is employed to obtain the global atmospheric light value of the hazy image, and finally, The atmospheric scattering model serves as a crucial tool when engaging in the operation of image dehazing, effectively enhancing the clarity of images affected by haze. The outcomes of the experiments indicate that this algorithm surpasses the latest approaches in both qualitative and quantitative evaluations. The algorithm put forward shows remarkable enhancements in peak signal - to - noise ratio and structural similarity, leading to significantlyenhanced dehazing performance.