Efficient Reflection Removal Algorithm for Single Image by Pixel Compensation and Detail Reconstruction

Gui-Cheng Tsai, Wei-Ting Chen, Shih-Yi Yuan, Sy-Yen Kuo

Abstract: When taking photos, there is often an undesired reflection image on glass windows, which may degrade the vision. In current reflection removal methods, they usually assume that the gradient of the transmitted layer is larger than the undesired refection layer and remove the object with smaller gradient. However, there is a challenging problem that these algorithms may remove the transmitted layer instead of the reflection layer when the assumption is violated. Moreover, some algorithms may not well preserve the detail and the reconstructed images are blurred. Furthermore, most of the existing algorithms require a lot of consumption time. Thus, we develop a new method based on pixel compensation and detail reconstruction to address these problems. By experimental results, the proposed method can enhance the transmitted layer effectively and the recovered images are clearer than other methods. In addition, the computation time is much less than that of prevailing methods.