Multi-Stage Based Feature Extraction Methods for Uyghur Handwriting Based Writer Identification

Abstract-Document image retrieval technology is extensively studied, but there is no report about Uyghur document image retrieval. Hu invariant moment features based document image retrieval scheme is proposed for Uyghur document images in this paper. Firstly, seven types of invariant moment features are extracted from Uyghur document images after obtaining image edge information using Canny edge operator. Then, the features are matched using Euclidean distance classifier and Feature distance classifier between query image with the target images. It is obtained the search results after ordering candidate images according to their similarity. Two types of experiments which are different in query image size are conducted using 1948 Uygur printed document images. The experimental results show that, the highest document image retrieval efficiency (to be get the matching rate of 100%) is obtained here when using the hole page document image is to be set as query image, and the matching rate is declined when sub images are selected as queries, the more the number of sub images, the lower the matching efficiency, and the retrieval efficiency is reached at minimum level when using one sixteenth of the document as query in our experiment. The experimental results indicated that Hu invariant moment features can effectively describe the nature of the Uyghur document images.

show abstract

“…At present, although studies on Hu moment invariants based image retrieval [8] is also increasing, ant it is used other aspects [11].…”

Section: Introductionmentioning

confidence: 99%