Color image or video typically has three color components: red(R), green(G) and blue(B), corresponding to the RGB color space. There are also many other color spaces such as YUV, YCbCr, YIQ, Lab, XYZ etc. For the common spaces YUV, YCbCr and YIQ, the color components are obtained mainly by performing a linear color transform on the RGB components. For a long time, researchers found that there exists strong intra-color and inter-color (or intra-channel) correlation in some aforementioned color spaces. For example, some researchers observed that local patterns in a image are often self-repeated locally/non-locally within the image. And some researchers observed that while image features such as edge or texture can be clearly observed in the R component, they are very often obse...[ Read more ]

Color image or video typically has three color components: red(R), green(G) and blue(B), corresponding to the RGB color space. There are also many other color spaces such as YUV, YCbCr, YIQ, Lab, XYZ etc. For the common spaces YUV, YCbCr and YIQ, the color components are obtained mainly by performing a linear color transform on the RGB components. For a long time, researchers found that there exists strong intra-color and inter-color (or intra-channel) correlation in some aforementioned color spaces. For example, some researchers observed that local patterns in a image are often self-repeated locally/non-locally within the image. And some researchers observed that while image features such as edge or texture can be clearly observed in the R component, they are very often observable in the G and B components as well. By utilizing the intrinsic intra- and inter-color correlation, we can improve some key techniques in High Efficiency Video Coding (HEVC) and image processing problems.

In this thesis, we address the problems of intra prediction and transform coefficients coding in video coding field, and problem of denoising the Bayer pattern color filter array (CFA) images. The first part of this thesis analyzes the chroma prediction mode, i.e. LM mode, and proves that the LM parameters for predicting original chroma and for predicting reconstructed chroma are statistically the same. We also analyze the error sensitivity of the LM parameters. We identify some LM mode problematic situations and propose three novel LM-like modes called LMA, LML and LMO to address the situations. To limit the increase in complexity due to the LM-like modes, we propose some fast algorithms with the help of some new cost functions. We further identify some potentially-problematic conditions in the parameter estimation (including regression dilution problem) and introduce a novel model correction technique to detect and correct those conditions.

Next, we investigated the inter-frame intra-color correlation and proposed a novel edge-based predictive scanning scheme to adaptively scan the transform coefficients from 2D array into 1D array in inter frames. We further proposed a technique for improving the coding efficiency of scanned transform coefficients using date hiding technique to hide multiple sign bits.

HEVC is mainly designed for natural content videos. With the increase of screen content video, HEVC Screen Content Coding (SCC) is developed as an extension of the H.265/HEVC. The third part of this thesis presents an independent uniform prediction (IUP) mode for improving the coding efficiency of screen content video. It’s used for both intra- and inter-coded blocks to exploit a set of globally occurring colors in a picture. At the coding unit level, IUP chooses one color out of a small set of global colors to form a uniform prediction block and then signals an index to indicate the selected color. All the pixels inside the coding unit can be predicted simultaneously.

Exploring intra- and inter-color correlation property in images also intrigue new solutions for image processing problems. The fourth part of this thesis proposes a joint denoising and demosaicking based on inter-color correlation (JDDC) scheme. This new framework explores the inter-color correlation between RGB channels. JDDC linearly combines an extracted luminance image and low-passed RGB images to get a full color image. Given the noise in the extracted luminance image and the low-passed RGB images are non-stationary and partially correlated, we modify the classical Non-Local Means (NLM) filter to denoise the extracted luminance image and the low-passed RGB iamges before the combination. The proposed new framework is both effective and computationally efficient.