| Summary: | 碩士 === 國立交通大學 === 資訊科學與工程研究所 === 101 === Screen contents are synthesized images generated by computer, which are generally composed by computer generated text, graphics, and natural images. Traditional hybrid video coding is inefficient to deal with such contents. To achieve a higher coding efficiency, this thesis classifies screen contents into text and pictorial pixels, and aims at comparing the performances of different methods for coding pictorial pixels. There are two kinds of pictorial pixels: one of which is computer graphics such as color gradient, graphic texture, and etc., and the other is natural images. The polynomial function is proposed mainly for dealing with color gradient while traditional discrete cosine transform is suitable for natural image. However, pictorial pixels cannot compose an entire image, and those non-pictorial pixels, which are viewed as holes, need to be filled before such incomplete image can be processed by discrete cosine transform. Other than that, this thesis also investigates other methods, such as matching pursuit and shape adaptive discrete cosine transform (SADCT), for incomplete images. The matching pursuit performs adequately for both color gradient and natural image, and is friendly for hardware design of the core transformation at decoder. However, the bases of matching pursuit may sometimes not suitable for characterizing the signal characteristics of graphic texture. A 2-D separable SADCT transform incorporates with variable-length 1-D DCT bases according to the horizontal and vertical lengths of input pixels. It will cause subjective discontinuity and will have distinct distortion on big fragment. The polynomial function works well on color gradient due to the capability of the polynomial function for generating similar gradient background. On the other hand, graphic texture works well at high rate by using palette-based scheme.
|
|---|
