Although recent deep learning-based algorithms have achieved significant performance improvements in various image enhancement tasks, most approaches have been developed using image-to-image translation, which is chal...
详细信息
Although recent deep learning-based algorithms have achieved significant performance improvements in various image enhancement tasks, most approaches have been developed using image-to-image translation, which is challenging to interpret and analyze the enhancement processes. Several attempts have been made to use the image-to-transformation function approach for better interpretability;however, they often fail to generate complex color mappings, degrading image quality. In this work, we develop a novel transformation function-based algorithm that estimates multiple transformation functions with different properties by exploiting both the spatial and statistical characteristics of the input image to describe complex color mapping. First, we extract the image features that capture spatial information, considering their channel correlations. Next, we estimate multiple transformation functions utilizing a cross-attention block to capture the relevance between spatial and statistical information in the input image and its histogram, respectively. We then estimate the weight maps indicating the pixel-wise contribution of each transformation function by exploiting the spatial correlation between the input and transformed images obtained by each transformation function. Finally, we obtain an enhanced image by taking the weighted sum of the transformed images and the corresponding weight maps. Experimental results demonstrate that the proposed algorithm outperforms state-of-the-art algorithms on various image enhancement tasks.
Most deep learning-based image enhancement algorithms have been developed based on the image-to image translation approach, in which enhancement processes are difficult to interpret. In this paper, we propose a novel ...
详细信息
Most deep learning-based image enhancement algorithms have been developed based on the image-to image translation approach, in which enhancement processes are difficult to interpret. In this paper, we propose a novel interpretable image enhancement algorithm that estimates multiple transformation functions to describe complex color mapping. First, we develop a histogram-based multipletransformation function estimation network (HMTF-Net) to estimate multiple transformation functions by exploiting both the spatial and statistical information of the input images. Second, we estimate pixel-wise weight maps, which indicate the contribution of each transformation function at each pixel, based on the local structures of the input image and the transformed images obtained by each transformation function. Finally, we obtain the enhanced image as the weighted sum of the transformed images using the estimated weight maps. Extensive experiments confirm the effectiveness of the proposed approach and demonstrate that the proposed algorithm outperforms state-of-the-art image enhancement algorithms for different image enhancement tasks.
暂无评论