Contrast Preserving Grayscale Method Based on Pseudo-Huber Loss

In response to the poor robustness of current image grayscale methods in handling outliers, this paper proposes a new contrast-preserving grayscale method based on Pseudo Huber loss. The method aims to improve the robustness of the algorithm based on maintaining the image contrast and maximally preserving the detailed features of the color image. The transformation model takes into account the fact that the Pseudo-Huber function is a combination of the robustness possessed by the MAE function and the continuous derivability of the MSE function, and as a result, a target energy function with Pseudo-Huber smoothing properties is constructed in the grayscaling model. Thus, the accuracy and color contrast of the detailed features are realistically maintained. To ensure the timeliness of the algorithm, the grayscaled discrete model is built using a discrete search solver. Experimental results show that the effectiveness of the proposed algorithm on the dataset proves the robustness of the method, and outperforms the state-of-the-art methods in both qualitative and quantitative metrics.