Texture for script identification

Abstract

Abstract-The problem of determining the script and language of a document image has a number of important applications in the field of document analysis, such as indexing and sorting of large collections of such images, or as a precursor to optical character recognition (OCR). In this paper, we investigate the use of texture as a tool for determining the script of a document image, based on the observation that text has a distinct visual texture. An experimental evaluation of a number of commonly used texture features is conducted on a newly created script database, providing a qualitative measure of which features are most appropriate for this task. Strategies for improving classification results in situations with limited training data and multiple font types are also proposed.

Automatic writer identification framework for online handwritten documents using character prototypes

“…Writer recognition systems can typically make use of global features such as texture, curvature and slant features [10][11][12] as well as a combination of local features such as graphemes, allographs and connected components [1,13,14] Our work falls into the latter category of text-independent techniques where the writers are not bounded by any specific lines of text in order for the system to recognize them. Instead, the system analyzes their handwriting styles through a series of automated processes, regardless of what they have written.…”

Segmentation of Tri-Lingual Documents

“…In contrast, global approaches (Joshi [4]) employ an analysis of regions (block of text) comprising atleast two lines (or words)without finer segmentation. In general, global approaches work well based on texture measurement, but this relies heavily on a uniform block of text (Buschet al [5]), and extensive preprocessing (to make the text block uniform) is required to measure the texture. Even though local approaches rely on the accuracy of character segmentation or connected component analysis, it could work well on the documents irrespective of their quality or uniformity in the block of text.…”

A Pixel Labeling Framework for Comparing Texture Features Application to Digitized Ancient Books

“…Fourteen textural features extracted of the GLCM have been initially introduced by (Haralick et al, 1973) for texture discrimination of natural and satellite images. A number of co-occurrence feature extraction and analysis methods (Mikkilineni et al, 2005;Lin et al, 2006;Payne et al, 1994;Peake and Tan, 1997;Busch et al, 2005) have been proposed in order to segment and classify the content of document images, and to identify script and language from document images. Briefly, the GLCM matrices are obtained for a small range of distance values d = 1, 2 and typically for the directions θ = {0 • , 45 • , 90 • , 135 • } (Busch et al, 2005).…”

“…Six co-occurrence features are extracted from the GLCM matrices: the maximum entry in the GLCM or the maximum probability, the correlation metric, the energy or the angular second moment, the entropy, the inertia or the contrast, and the local homogeneity for two distances {d = 1, 2} (Mikkilineni et al, 2005;Busch et al, 2005). In addition to the twelve cooccurrence features (six for each distance), two other descriptors are computed: the mean value and the standard deviation of the energy for the two distances combined (Lin et al, 2006).…”

“…Several statistical texture-based segmentation methods have been presented, e.g. the autocorrelation function (Petrou and Sevilla, 2006) and GLCM (Grey Level Co-occurrence Matrix) (Haralick et al, 1973;Busch et al, 2005). The extracted texture features are mainly investigated and analyzed separately in independent experiments for document analysis (Journet et al, 2008).…”