Browse Prior Art Database

Fixed Set of Masks for Feature Extraction Measurements

IP.com Disclosure Number: IPCOM000083688D
Original Publication Date: 1975-Jul-01
Included in the Prior Art Database: 2005-Mar-01
Document File: 2 page(s) / 60K

Publishing Venue

IBM

Related People

Stafford, SK: AUTHOR

Abstract

In known feature extraction processes of character recognition, before each measurement is applied, a mask of binary 1's and 0's must be loaded into the logic unit accumulator (LUA). The mask preconditions the LUA to cause data in line with the mask 0 bits to be ignored and data in line with the mask 1 bits to be considered significant. In this way, the pattern of 1 bits in the mask defines the size and shape of a window that exactly fits the requirements for this one measurement. It is through this window that the measurement will be taken or applied.

This text was extracted from a PDF file.
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.

Page 1 of 2

Fixed Set of Masks for Feature Extraction Measurements

In known feature extraction processes of character recognition, before each measurement is applied, a mask of binary 1's and 0's must be loaded into the logic unit accumulator (LUA). The mask preconditions the LUA to cause data in line with the mask 0 bits to be ignored and data in line with the mask 1 bits to be considered significant. In this way, the pattern of 1 bits in the mask defines the size and shape of a window that exactly fits the requirements for this one measurement. It is through this window that the measurement will be taken or applied.

The masks could be generated by the same program that compiles the measurement microcode. One approach is to have this program generate a separate mask for every window used. Using 64-bit windows, this would yield 279 masks and for 32-bit windows, 517 masks. However, many of these 279 or 517 masks will be duplicates.

To continue producing masks to fit each measurement as it was compiled, but to avoid producing duplicates, would require keeping track of the dimensions of every unique mask generated and its sequentially-assigned-as-needed storage location. Then the program could check the list to see if the needed mask had already been produced. If it had been produced, the program could point the load-mask instruction to the storage location of that mask. If not, it could then create a new mask.

This approach could very well produce less than 65 unique masks, be...