AWB Algorithm Based on Improved U-Net Network

This paper proposes an automatic image white balance method based on enhancing the U-Net network, addressing issues of color cast resulting from improper white balance settings. By incorporating channel attention mechanisms prior to downsampling within the U-Net architecture, each channel of the input image undergoes weighted processing. The resultant weighted feature maps contain richer and more precise detail, thereby enhancing color correction processes. Experimental outcomes demonstrate that images corrected using this algorithm exhibit strong color constancy, with both subjective and objective evaluations yielding excellent results.