2007 Fiscal Year Final Research Report Summary
Fast Retrieval of Massive linage Archives based on Similarity of I cleat Feature Descriptor
| Project/Area Number |
18300061
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Perception information processing/Intelligent robotics
|
|---|
| Research Institution | Future University-Hakodate |
|---|
Principal Investigator |
KAWASHIMA Toshio Future University-Hakodate, SCHOOL OF SYSTEM INFORMATION SCIENCE, PROFESSOR (20152952)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
NAGASAKI Takeshi FUTURE UNIVERSITY-HAKODATE, SCHOOL OF SYSTEM INFORMATION SCIENCE, ASSOC. PROFESSOR (70325893)
|
|---|
| Project Period (FY) |
2006 – 2007
|
| Keywords | ARCHIVES / IMAGE RETRIEVAL / LOCAL DESCRIPTOR / HASHING |
|---|
| Research Abstract |
In the research projects we developed fast retrieval method of historical document database using local feature of images. We also developed codes for local feature detection for MX-1 parallel processors, and LSH algorithm for fast character recognition. The followings are the outcomes of the project.
1 Historical document retrieval based on PCA
Eigenspace method (PCA) applied to document image matching showed fast retrieval of document database. Image of character strings are sliced into slit-wise small images and reduced the dimension using eigenspace method. The performance of retrieval for was 73-93 percent.
2 Retrieval using gradient histogram
We refined feature vector based on PCA vector to gradient histogram. The approach improved the performance of PCA to 95-98 percent.
3 Application of local feature matching
We developed a transcript mapping software for historical document based on the pattern of frequency of words in the document using method 1.
4 Fast computation of local feature using parallel processor
Algorithms of SURF image feature are implemented on a MX-1 parallel processor. Computation of scale detection and gradient distribution calculation are accelerated using the processor.
5 Locality sensitive hashing of local feature vector
We discussed fast dictionary retrieval method for reproduction of historical document. We implemented Locality Sensitive Hashing algorithm for document image retrieval to search similar character image from dictionary of character image examples.
|
Research Products
(10 results)
