數位影像穩定系統之新演算法與架構

Abstract

在這篇論文中，我們針對數位攝影機之數位影像穩定系統提出一種新的演算法與架構。這個方法是利用灰碼位元平面區塊比對法來做晃動估測以消除攝影機非故意之晃動所造成攝得影像產生令人不舒服的感覺。在我們的方法裡，使用灰碼位元平面區塊比對法可以降低運算的複雜度。為了改善晃動估測的性能，動態影像裡的每一張畫面被區分成許多區塊以做區域性之晃動偵測。根據我們的架構，晃動評定單元利用時間上的關連性可以有效地偵測影像裡具有移動物體或攝影機故意之移動等情況。在晃動補償單元我們使用仿射運動模型以補償攝影機之轉動晃動，同時並不會增加太多的運算。根據所提出的架構，本文也同時設計一個即時晃動偵測器。為了程式彈性化與硬體效率化的考量我們將晃動評定單元與晃動補償單元程式化寫入一個連接晃動偵測器的微處理器。本文所提之影像穩定器現階段正在實現於一塊FPGA的板子上。

In this thesis, we propose a new algorithm and architecture to do image stabilization for digital camcorder. This approach is based on gray-code bit-plane block matching to eliminate the unpleasing effect caused by involuntary hand movement of the camera holder. In our approach, gray-code bit-plane block matching is used to reduce the computation complexity. To improve the performance of stabilization, each frame is divided into several blocks to do localized motion estimation. Based on our architecture, the temporal correlation is utilized at the Motion Decision Unit to efficiently detect moving objects and intentional panning. To compensate for camera rotation, affine motion model is used at Motion Compensation Unit without adding too much computation load. A real-time Motion Estimator is also proposed based on the proposed architecture. Having considered both programming flexibility and hardware efficiency, the Motion Decision Unit and Motion Compensation Unit are coded in a microprocessor that interconnects with the Motion Estimator. The proposed stabilizer is now implemented on an FPGA board. Chapter 1 Introduction…………………………………………… 1 1.1 Digital Image Stabilization for Camcorders………………………...… 1 1.2 System Overview……………………………………………………… 2 1.3 Outline………………………………………………………………… 3 Chapter 2 Backgrounds..……...………………………………….. 4 2.1 Motion Estimation…….……………………………………………… 4 2.1.1 Pre-processing of Motion Estimation………………………….. 4 2.1.2 Block Matching……………………………………………….. 5 2.1.3 Projection Matching and Other Matching Methods….……….. 7 2.2 Motion Decision…….………………………………………………… 8 2.2.1 Interfering Factors………………………...…………………… 8 2.2.2 Previously Proposed Methods for Motion Decision…………... 9 2.3 Motion Compensation………………………………………………… 10 2.4 Image Interpolation…………………………………………………… 12 Chapter 3 Multi-Resolution Block Matching over Gray-Code Bit-Planes………………………………………...……. 13 3.0 Basic Architecture…………………………………………………….. 13 3.1 Block Matching over Bit-Planes…………………….………………… 14 3.2 Block Matching over Gray-Code Bit-Planes……………………….… 15 3.3 Multi-Resolution Block Matching……………….……………………. 22 3.4 Motion Estimation Architecture…...…………….……………………. 24 Chapter 4 Motion Decision and Motion Compensation………... 26 4.1 Motion Decision………………………………………………………. 26 4.1.1 Lack-of-Feature Condition…………………………………….. 27 4.1.2 Existence of Moving Objects………………………………….. 29 4.1.3 Intentional Panning Condition…………………………………. 31 4.1.4 Existing Repeated Patterns Condition………….…...…………. 32 4.1.5 Optical Zooming Condition……………………………………. 32 4.1.6 Spatial Noise Checking of Noise Level…………………….…. 33 4.1.7 Procedure of Motion Decision……...…………………………. 34 4.2 Motion Compensation…………...……………………………………. 35 4.2.1 Translational Motion Model……………...……………………. 35 4.2.2 Affine Motion Model………...…...…………………………… 37 4.3 Digital Zooming…………….………………………………………… 40 Chapter 5 Hardware Implementation for the DIS System……... 42 5.1 Real-Time Motion Estimation………………………………………… 43 5.2 Overview of Motion Estimator………..………………………………. 45 5.3 Details of Motion Estimator………….……………………………….. 47 5.3.1 Preprocessor………….………...…...…………………………. 48 5.3.2 Correlation Unit………………...…...…………………………. 49 5.3.3 Correlation Memory with Its Controller………………………. 54 5.3.4 Bit-Plane Memory and Bit-Plane Memory Controller……..…. 55 5.3.5 Comparison Unit…………………………………….….…..…. 56 Chapter 6 Conclusions…….…………………………………….... 60 Bibliography 61