Browsing by Author "Kuo, IS"

Now showing 1 - 2 of 2

High visual quality sprite generator using automatic segmentation and intelligent blending
(2005)
Kuo, IS
;
Chen, LH
;
資訊工程學系
;
Department of Computer Science
This paper presents a sprite generator using automatic segmentation and intelligent blending. Unlike the sprite generator introduced in MPEG-4, which requires manually segmentation masks provided by user, the proposed sprite generator segments a video frame automatically into the background and moving objects called masks first. The segmentation masks increase the fidelity of the generated sprite. The automatically segmented masks are then used to generate sprite. The experimental result shows that the generated sprite using the automatically segmented masks close to the generated sprite using manually segmented masks. An intelligent blending method is also proposed to remove ghostlike shadows in the generated sprite caused by imperfect segmentation. The intelligent blending removes the shadows effectively without PSNR degradation, and the fidelity of the generated sprite is raised.
Predictive classifier for image vector quantization
(2000-09-01)
Tai, SC
;
Wu, YG
;
Kuo, IS
;
資訊工程學系
;
Department of Computer Science
A new scheme for a still image encoder using vector quantization (VQ) is proposed. The new method classifies the block into a suitable class and predicts both the classification type and the index information. To achieve better performance, the encoder decomposes images into smooth and edge areas by a simple method. Then, it encodes the two kinds of region using different algorithms to promote the compression efficiency. Mean-removed VQ (MRVQ) with block sizes 8 x8 and 16x16 pixels compress the smooth areas at high compression ratios. A predictive classification VQ (CVQ) with 32 classes is applied to the edge areas to reduce the bit rate further. The proposed prediction method achieves an accuracy ratio of about 50% when applied to the prediction of 32 edge classes. Simulation demonstrates its efficiency in terms of bit rate reduction and quality preservation. When the proposed encoding scheme is applied to compress the "Lena" image, it achieves the bit rate of 0.219 bpp with the peak SNR (PSNR) of 30.59 dB. (C) 2000 Society of Photo-Optical Instrumentation Engineers. [S0091-3286(00)00908-9].