Logo image
適用於低位元率編碼之可形變網狀動態影像補償與壓縮
Thesis

適用於低位元率編碼之可形變網狀動態影像補償與壓縮

劉泰誠
Masters, 國立清華大學, 電機工程學系
1996

Abstract

運動估測 運動補償 控制點內插 紋路映射 仿形轉換 雙線性轉換 Motion Estimation Motion Compensation Control Grid Interpolation Texture Mapping Affine Transform Bilinear Transform
在本篇論文中, 我們將介紹一種以三角形區塊為主的運動估測與 補償的方法來克服一些以方塊為主的傳統影像編碼器所遭遇到的一些缺 點, 如方塊效應或飛蚊效應。為了克服上述的兩個缺點, 我們首先將影像 分割成數個三角區塊, 再分別對每個區塊做運動估測與補償。此外, 這個 心的運動估測與補償方法也能處理影像擴大與縮小(zoom)和旋轉效應。 同時, 為了提高壓縮率, 我們必須對影像做分割(segment),以減少控制點 的數目。在這篇論文中, 我們採用四元樹的結構來實現。並同時採用以運 動為主的基準來分割。稍後, 我們也結合多重解析度的觀念來加速運動估 測。此外, 我們也提出一個選擇性改善的方法, 將我們電腦運算的時間盡 量集中在一些比較容易出錯的控制點上。 最後, 將我們的網狀結 構應用到以物件為導向的MPEG-4影像序列上。 A new triangle-based motion estimation and motion compensation method is proposed in this thesis to overcome the drawbacks of conventional block-based codecs. To avoid the blocking artifacts and mosquito effects,we partition the image frame into several triangles and then calculate their affine transformation parameters to complete the motion estimation.The motion compensation is done by generic texture mapping method in computer graphics. Besides, this new motion estimation can estimate thezooming and rotational effects which are not dealt with by generic blockmatching algorithm. To achieve high compression ratio, we must segment our image framefirst to reduce the control grid points. In this thesis, we use quadtree segmentation method with motion-based criterion. Later, we combine the multi-resolution concept with our Control Grid Interpolation(CGI) methodto replace block-matching method in the motion estimation step. Besides, we also propose a selective refinement algorithm to concentrate computer calculations on large-error grids. Finally, we adopt the concept of content- based approach in MPEG-4 in our mesh structure and compare our proposed method with H.263-TMN5 in both the reconstructed PSNR and coded bits.

Metrics

1 Record Views

Details

Logo image