Interference reflection separation from a single image

Separation of weak reflection from a single superimposed image using gradient profile sharpness

“…We compare our method with Kayabol et al's method [9] and Chung et al's method [10]. These two methods also realize the separation of a single superimposed image automatically.…”

“…As compared with the most related previous work of [10], the proposed algorithm can obtain greatly improved performance by adopting two effective approaches. Firstly, we adopt GPS (gradient profile sharpness) [11] to represent edge smoothness.…”

“…There always exist a few tone distortions in its separated results. Recently, Chung et al discovered the phenomenon that the edges of weak reflection were quite smooth [10]. They introduced the smoothness constraint into the classification of gradients in the superimposed image.…”

“…Thus it can be assumed that the significant gradients in S I are influenced by only one component image ( O I or I I ). This assumption has been adopted in most previous separation methods [2,6,8,10].…”

Reflections Removal Using K-Means Clustering

“…Remove reflection from a single image is a highly difficult problem compared with multiple image reflection removal since trying to obtain two images (reflection image and a transmission (background) image) from just a single superimposed image is a hard problem but practical scenario and a small number of aspirant methods try to solve it [9]. Some existing works are used sparse gradient priors (a sparsity prior stated that the output of any derivative filter tends to be sparse where the histogram of the output of a derivative filter is peaked at zero and fall off rapidly out to the two extreme ends of the histogram) to classify the reflection and transmission edges as in [10,11] that depend on user intervention to mark the reflection and transmission layer, authors in [12] utilized values of gradient in direct manner, where separated images are reconstructed from the classified gradients that based on the smoothness constraint in the classification of gradients in the superimposed image, as well in [13] that based on the gradient prior to separate the two layers by imposing a sparse gradient prior in the transmission layer and a smooth gradient prior on the reflection layer, the work of [14] is tried to exploit ghosting artifacts that are typical of images captured through a window, to separate the layers the GMM (Gaussian Mixture Model) for regularization is used in the proposed algorithm. Other methods used (Depth Of Field) DOF map to classify edges to reflection and background as in [1].…”