Browse Prior Art Database

Arbitrary Anti-aliasing Scale & Rotation Based on index-tables

IP.com Disclosure Number: IPCOM000010197D
Original Publication Date: 2002-Nov-04
Included in the Prior Art Database: 2002-Nov-04
Document File: 4 page(s) / 149K

Publishing Venue

Motorola

Related People

David Hayner: AUTHOR [+3]

Abstract

Even if scaling and rotation are very common operations in many digital imaging, due to the computational complexity, Hardware implementations of arbitrary anti-aliasing scaling and rotation has been very rarely challenged. The authors propose an efficient and arbitrary anti-aliasing scaling and ration scheme based on index-table. By using this scheme, 1 degree rotation and 1% scaling can be implemented in hardware very efficiently. Another advantage of this scheme is that scaling and rotation are analyzed from a unified viewpoint so that the two operations can be easily merged in an efficient implementation.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 50% of the total text.

Arbitrary Anti-aliasing Scale & Rotation

Based on index-tables

David Hayner

Sang Park

Reginald O’Donoghue

Abstract

Even if scaling and rotation are very common operations in many digital imaging, due to the computational complexity, Hardware implementations of arbitrary anti-aliasing scaling and rotation has been very rarely challenged. The authors propose an efficient and arbitrary anti-aliasing scaling and ration scheme based on index-table. By using this scheme, 1 degree rotation and 1% scaling can be implemented in hardware very efficiently. Another advantage of this scheme is that scaling and rotation are analyzed from a unified viewpoint so that the two operations can be easily merged in an efficient implementation.

1.       Scaling

The classic approach to scaling involves upsampling by some factor, filtering, and then downsampling [1].� This approach requires storing upsampled data that may not be used in downsampling process.� By selecting a filter length that performs reasonably well for a variety of scaling factors as well as carefully designing look up tables we can process the exact data in the upsampling block that needs to be reused in downsampling.� Further computational gains can be achieved by dithering.� Dithering effectively allows us to get 1% resolution based on 5% offset resolution.� For example to get a 47% scaled image we skillfully modulate between a 50% and 45% scaled outputs. A closed form of scaling is provided in Equation 1. Based on Equation 1, tables can be efficiently generated for each resolution on the fly. For example, to achieve 20% to 200% scaling with 1% offset, about thirty tables will be used. Since the size of each typical table size is about 100 bytes, about 3KB memory space would be necessary to save all tables to support scaling between 20% and 200% with 1% offset. Or each table can be also dynamically generated for give specific scaling input. The contents of those tables are the exact index of both input and filter to produce scaled outputs. Therefore, all scaling processing can be done by simply table look-up and MAC (multiply accumulate unit) operation without unnecessary data movement or buffering to achieve arbitrary scaling. Figure 1 demonstrates the basic difference between a typical approach and the method proposed here. For instance, to achieve 86% (P=86) scaling, by using L = 17, only required inputs and filter coefficients will be assigned by the tab...