PhotoGraphic Technology Day 24: Quantization, image texture, and superresolution скачать в хорошем качестве

PhotoGraphic Technology Day 24: Quantization, image texture, and superresolution

Google Tech Talks August 1, 2007 ABSTRACT Quantization is an intrinsic feature of digital signals, an unavoidable nonlinearity in representation. It's instructive to analyze scalar quantization as part of the signal sampling process, how it characterizes the signal, and how signals so characterized can best be reconstructed. The resulting process has many advantages over conventional interpolation. It can reconstruct image structures such as edges and contours with sub-sample precision, an important ingredient of "superresolution." An open question is whether analogous techniques have a role to play in vector quantization. In images, vector quantization is often used spatially, to encode groups of pixels at a time. Typically this is used for data compression. Recent work in texture synthesis, however, appeals to vector quantization to characterize and reproduce the joint statistics of pixel values in a "random" field. In current usage, this is an optimization for what would otherwise be a big search; vector quantization is used to index spatial blocks of "texture." In an early effort in this area, vector quantization was used to synthesize texture by capturing correlations among spatial VQ blocks at different scales. This idea leads to a characterization and interpolation of texture that delivers fine detail at all scales, in the manner of a fractal. After a very brief review of leading texture synthesis techniques, some recent work in continuous scaling of image textures will be demonstrated, and the prospects for combining the superresolution of image structures and image textures will be discussed. useful properties of light and image formation theory and techniques of photographic optics and image capture theory of colorimetry and techniques of color reproduction where and how photography is being used in Google products and projects what tools exist inside Google for photographic image storage, processing, etc. and lots more... Google engEDU Speaker: Lance Williams Speaker: Diego Rother