Blind Visual Motif Removal From a Single Image

Abstract

Figure 1: Blind visual motif removal results on images unseen during training. Top: test images embedded with semitransparent motifs. Bottom: our reconstructed results. Our network was trained on Latin characters, yet successfully identifies and removes the Hindi and Japanese characters (left three images). Similarly, the overlaid visual motifs on the right three images differ semantically from the motifs used during training. AbstractMany images shared over the web include overlaid objects, or visual motifs, such as text, symbols or drawings, which add a description or decoration to the image. For example, decorative text that specifies where the image was taken, repeatedly appears across a variety of different images. Often, the reoccurring visual motif, is semantically similar, yet, differs in location, style and content (e.g., text placement, font and letters). This work proposes a deep learning based technique for blind removal of such objects. In the blind setting, the location and exact geometry of the motif are unknown. Our approach simultaneously estimates which pixels contain the visual motif, and synthesizes the underlying latent image. It is applied to a single input image, without any user assistance in specifying the location of the motif, achieving state-of-the-art results for blind removal of both opaque and semi-transparent visual motifs.We present a method for completely blind visual motif removal. In the blind setting, the exact location, structure and size of these motifs is unknown. The generalization ability of our network is demonstrated by removing visual motifs that are not seen during training, e.g., watermark re-

Visible Watermark Removal via Self-calibrated Localization and Background Refinement

“…As exemplified in Figure 2, our whole network is designed in a coarse-to-fine manner, which comprises of a coarse stage and a refinement stage. In the coarse stage, similar to previous multi-task learning methods [19,28], we employ one shared encoder and two split decoders, in which two decoders account for localizing the watermark (mask decoder branch) and restoring the background image (background decoder branch) respectively. In the mask decoder branch, we design a Selfcalibrated Mask Refinement (SMR) module to promote the quality of predicted watermark mask.…”

“…Although these methods are capable of merging multi-level features, how to propagate multi-level information properly and efficiently in watermark removal task is still unsolved. In watermark removal approaches [19,28], Hertz et al [19] only considered the skip connection from encoder; Liu et al [28] further passed the shallowest decoder feature from coarse stage to refinement stage. Nevertheless, these methods overlook the potential capacity of multi-level features integration.…”

“…In the coarse stage, we adopt the U-Net [33] architecture with skip links connecting encoder and decoder features as shown in Figure 2. Specifically, we employ the structure of encoder block and decoder block in [19]. Watermark localization and watermark removal are treated as two tasks, which share all five encoder blocks and the first decoder block.…”

Image Inpainting: A Review

“…The same goal is reached by Cai et al [41] who proposed a semantic object removal approach using CNN architecture. In order to remove motifs from single images, Hertz et al [42] proposed a CNN-based approach. Table 2 summarizes CNN-based method with a description of the type of data used for image inpainting.…”

“…For example, the images damaged by blocks are less accurate in term of PSNR values. The algorithms[29,31,39,42] can handle the added visual motifs like text or lines with a good performance. In addition, the performance is influenced by the percentage of added noise to the images.…”