Fast Single Image Reflection Suppression via Convex Optimization

Abstract

Removing undesired reflections from images taken through the glass is of great importance in computer vision. It serves as a means to enhance the image quality for aesthetic purposes as well as to preprocess images in machine learning and pattern recognition applications. We propose a convex model to suppress the reflection from a single input image. Our model implies a partial differential equation with gradient thresholding, which is solved efficiently using Discrete Cosine Transform. Extensive experiments on synthetic and real-world images demonstrate that our approach achieves desirable reflection suppression results and dramatically reduces the execution time.

Single Image Reflection Removal Based on GAN With Gradient Constraint

“…Recently, some methods were proposed to remove reflection without using multiple images [9]- [17]. Methods in [9]- [11] are based on solving optimization problem. Since understanding the image structure is important in removing reflections, Convolutional Neural Network (CNN) was applied in some methods [12]- [14].…”

Single image reflection removal with camera calibration

“…To make this problem more tractable, several priors have been proposed in the literature. Some of them are based on the physical form of the reflection, e.g., ghost effects due to double surfaces of glass [40], or blurry effect due to out-of-focus and limited depth of field of the camera [30,31,62]. Some other priors are observed from transmission and reflection, e.g., the sparsity of gradient due to natural image priors [29,2], or image contents priors due to known scenes or objects [46].…”

“…Wan et al [46] employ both content and gradient priors to jointly restore missing contents and recover transmission images [46]. Yang et al [62] suppress the reflection by solving a partial differential equation.…”

“…Zoom in for better details. As can be observed, ZN18 [64] and WS19 [49] produce darker results which could be seen as color distortion, YM19 [62] and WY19 [52] output over-smooth ones which could destruct the original background image for further use, and results from WT19 [53] contain unexpected artifacts. Our method alleviates these problems and hence produces better results as compared with other methods.…”

“…Specifically, we compare the performance with five state-of-the-art single image reflection removal methods, including an optimization-based method (i.e., YM19 [62])…”