Image Reflection Removal Method Based on Edge Clues and Convolutional Guided Filters

When using devices such as mobile phones and cameras to take photos of target objects through glass, the glass often reflects the image of the object that interferes with the lens. Due to the fact that reflection images and transmission images are both objects that exist in the real world, there is a high degree of feature similarity in their structure, texture, and other aspects. As a result, the separated transmission images often contain residual reflections, resulting in varying degrees of edge loss and texture distortion in transmission images. In response to these issues, this article proposes an image reflection removal method based on edge clues and convolutional guided filtering. Provide powerful guidance for restoring better transmission images using flash-only images; Input edge clues based on gradient mapping into the network, guiding the network to correctly predict the transmission layer; Using convolutional guided filters to combine the original image, the first stage predicted image, and the edge information of the image, the final predicted result is comprehensively guided and outputted. The performance of the model was evaluated on the FRRD dataset, and the results showed that the model had good reflection removal performance while maintaining image color and edge distortion, which was superior to other advanced algorithms.