Number Of Example Images Research Articles

Deep Image Prior

Deep convolutional networks have become a popular tool for image generation and restoration. Generally, their excellent performance is imputed to their ability to learn realistic image priors from a large number of example images. In this paper, we show that, on the contrary, the structure of a generator network is sufficient to capture a great deal of low-level image statistics prior to any learning. In order to do so, we show that a randomly-initialized neural network can be used as a handcrafted prior with excellent results in standard inverse problems such as denoising, super-resolution, and inpainting. Furthermore, the same prior can be used to invert deep neural representations to diagnose them, and to restore images based on flash-no flash input pairs. Apart from its diverse applications, our approach highlights the inductive bias captured by standard generator network architectures. It also bridges the gap between two very popular families of image restoration methods: learning-based methods using deep convolutional networks and learning-free methods based on handcrafted image priors such as self-similarity. Code and supplementary material are available at https://dmitryulyanov.github.io/deep_image_prior .

A VQ-DCT Scheme for Grayscale Image Watermarking

paper, a hybrid watermarking scheme, that employs both DCT and VQ, has been presented. The authors have proposed a modification of the conventional LBG algorithm in this paper for designing the codebook to be used subsequently for compressing the image employing VQ. The watermark/logo is embedded in the constructed codebook after suitable DCT operations. The compression and reconstruction of the cover image employs the modified embedded codebook. The indirect mechanism of embedding the watermark and the hybrid combination of VQ and DCT make the entire watermarking scheme reasonably robust with respect to various types of common attacks which include intensity modification in real domain, modification in transformed domain namely JPEG compression etc. The method has been implemented and tested on a number of example images. We have also tested the robustness of authenticating capability of the scheme both qualitatively and quantitatively for various types of attacks.

A simple light simulation algorithm for massively parallel machines

A new computer graphics algorithm which simulates the propagation of light and its interaction with matter on a massively parallel machine is presented. This algorithm, called the Tagged Shooting Method, is designed for a virtual computer containing a great number of simple, communicating processors arrayed into a cubical, three-dimensional lattice. Only nearest-neighbor communication among processors is assumed and there is no reliance on global shared memory. The algorithm is similar in spirit to the classical Progressive Refinement Radiosity Method designed for more conventional computers but is not an adaptation of that technique to massive parallelism. Instead, the new algorithm uses a discretization of the wave equation as a local rule for shuttling radiant energy values between processors which correspond to regions of space. A number of example images that were created with an implementation of the algorithm on a Connection Machine are depicted and critiqued.