Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Image Segmentation Algorithm and Implementation for a Programmable Media Processor

IP.com Disclosure Number: IPCOM000021739D
Publication Date: 2004-Feb-05
Document File: 6 page(s) / 2M

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method for an image segmentation algorithm that effectively separates text from the graphics, images, and/or background of a document; the disclosed method segments documents into different compression domains, enhancing the overall data compression without sacrificing output document quality.

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 53% of the total text.

Image Segmentation Algorithm and Implementation for a Programmable Media Processor

Disclosed is a method for an image segmentation algorithm that effectively separates text from the graphics, images, and/or background of a document; the disclosed method segments documents into different compression domains, enhancing the overall data compression without sacrificing output document quality.

Background

In general, segmentation is the process used to distinguish between the objects of interest in an image (sometimes referred to as “foreground”), and the rest of the image (sometimes referred to as “background”). In document processing, it is common to separate texts from the graphics, images, and/or the background of a document. This separation allows effective processing of each segmented part of the document, without sacrificing output document quality. Once separation of text or graphics from the background is achieved, one can process the text through lossless compression (like G4) while the images and background are processed through lossy compression (like JPEG). There is no segmentation technique that works for all images, and no segmentation technique is perfect. The two most common techniques used are “thresholding” and “edge finding”.

General Description

The disclosed method uses a pixel based “thresholding” technique, consisting of two steps:

1.      Using the variance of a surrounding 3x3 block of luminance channel data of the document image.

2.      Adding an additional post processing step to improve the quality of segmentation without degrading the performance of the algorithm: the amount of variance change in the 3x3 variance map is computed by measuring the maximum and the minimum of variance distribution in the surrounding 3x3 variance neighborhood.

The disclosed method uses two MXP Image Signal Processors: one for the first step (computation of initial variance), and one for the second step (computation of the change in variance). The configurability and programmability of the MXP5800 architecture allows one to implement the disclosed method’s segmentation algorithm either with only one ISP (skipping the post processing done by the second ISP) to save the ISP resource, or implement with two ISPs for better overall segmentation results.

The main components of the MXP5800 used for segmentation are the MCH unit and two MAC-PE units inside an ISP. These are used for efficient computation of the 3x3 block variance, and...