A Fine-Granularity Down-Sampling Circuit for Image Pyramid Applications
| Autor: | Wang, Chiang-Yi, 王強毅 |
|---|---|
| Rok vydání: | 2018 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 106 Image scaling (image scaling) has been widely used in the multimedia field, and the target of this thesis is to design a sampling circuit with a fine decimation rate, providing a fine 24-level sampling ratio (decimation rate), and the common implementation of image pyramid providing only 3 to 4 -level [10][11][12]. Proposed down scaler has greater flexibility for different resolutions. It can be applied not only to the image pyramid [1] that produces multi-resolution, and to the training data required for the image recognition of CNN, which can produce more different resolution sizes as data, so that it has a better recognition effect[4]. Common software image scaling algorithms, such as bilinear interpolation and bicubic interpolation, are all non-adaptive, meaning that any decimation rate, all pixel scaling algorithms are the same. In this paper, we adopted that the Sine-windowed Sinc-function down-sampling filter[5] mentioned in H.264/SVC is adaptive for decimation rate, which can be adjusted for different decimation rate. A lower resolution layer is generated by the down scaling. A series of image sets are generated and the image details are retained. Using the image pyramid that keeps the details of the image and providing multi-resolution with the fast speed of the circuit, the future can be applied to approximate the non-linear operation to the linear operation, such as the use of the optical flow[2][3] calculation. At present, we proposed down scaler architecture, which produces 8-level image pyramid at most in real-time (640x480@30fps). |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|