DOCUMENT IMAGE SEGMENTATION SYSTEM

• US 20100290701A1
• Filed: 04/01/2010
• Published: 11/18/2010
• Est. Priority Date: 05/13/2009
• Status: Active Grant

• Current Assignee: TATA Consultancy Services Limited (Tata Sons Pvt Ltd.)
• Original Assignee: TATA Consultancy Services Limited (Tata Sons Pvt Ltd.)
• Inventors: Hiranmay, Ghosh, Puneet, Jeff, Ashish, Khare
• Granted Patent: US 8,320,665 B2
• Time in Patent Office: Days
• US Class Current: 382/164
• CPC Class Codes: G06T 2207/10024 Color image
• CPC Class Codes: G06T 2207/20036 Morphological image processing
• CPC Class Codes: G06T 2207/30176 Document
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/155 involving morphological operators
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/162 Quantising the image signal
• CPC Class Codes: G06V 30/18105 related to colour
• CPC Class Codes: G06V 30/192 using simultaneous comparisons or correlations of the image signals with a plurality of references

Abstract

A system and a method for document image segmentation have been disclosed. Image segments are obtained by forming different clusters in a document image. The document image may include images of company logos, product marks or trademarks. The invention can perform image segmentation on any kind of complex coloured image and can recognize logos, product-marks or trademarks which comprise text or graphics, wherein the text can be either of uniform font style or uneven font style such as fancy font styles, calligraphic styles or having different orientation.

Claims

1. A system for document image segmentation, said system comprising:

input means adapted to input a document image;

image pre-processing means adapted to pre-process said document image by maintaining the aspect ratio, said pre-processing means including a colour quantization means to give a pre-processed quantized image;

colour space transformation means adapted to receive said pre-processed quantized image and apply a Hue, Saturation and Value colour space transformation on said quantized image to derive a transformed image containing only saturation component of said quantized image;

first image energy calculation means adapted to receive said transformed image and calculate both horizontal and vertical energies of said transformed image to provide a first energy image by cumulating both of said calculated energies of said transformed image;

grayscale image conversion means adapted to receive said pre-processed quantized image and perform a grayscale conversion operation on said quantized image to provide a gray scale image;

second image energy calculation means adapted to receive said gray scale image and calculate both horizontal and vertical energies of said gray scale image to provide a second energy image by cumulating both of said calculated energies and said gray scale image;

computational means adapted to receive said first energy image and second energy image to compute a maximum of both the energies and provide a maximum energy image;

binarization means adapted to receive said maximum energy image and provide a binarized image;

dilation means adapted to receive said binarized image and perform a dilation operation to provide a dilated image;

clustering means adapted to receive said dilated image and formulate different clusters based on the density of the dilated areas and provide a clustered image; and

box creation means adapted to create bounding boxes enclosing each cluster in the clustered image to form an image of the document having image segments.

2. The system as claimed in claim 1, wherein said system includes an document image analyzing unit, said analyzing unit comprising:

a database containing images of logos, product-marks and trademarks along with company and product information of a plurality of companies;

indexing means adapted to index said images of logos, product-marks and trademarks stored in said database;

receiving means adapted to receive different image segments of a document for analysis of the document;

matching means adapted to match said different image segments with said images of logos, product-marks and trademarks stored in said database to provide matching image segments;

extraction means adapted to receive said matching image segments and extract the company and product information corresponding to each of said matching image segments; and

appending means co-operating with said matching means adapted to append non matching image segments containing new logo, product-mark or trademark into the database, together with company or product information.

3. The system as claimed in claim 1, wherein each pixel of said quantized image is represented by four bits.

4. The system as claimed in claim 1, wherein said dilation means is adapted to add pixels at boundaries of black pixels in the binarized image thereby providing a thickened and connected image.

5. The system as claimed in claim 1, wherein said image pre-processing means also includes a noise removing means.

6. The system as claimed in claim 1, wherein said clustering means is adapted to use DBSCAN technique to compute clusters.

7. A method for document image segmentation, said method comprising:

inputting a document image;

pre-processing input image by maintaining the aspect ratio and performing colour quantization to give a quantized image;

applying Hue, Saturation and Value colour space transformation on said quantized image;

deriving a transformed image containing only saturation component of the quantized image;

calculating both horizontal and vertical energies of said transformed image and cumulating both of the calculated energies;

providing a first energy image;

converting a quantized image into a grayscale image;

calculating both horizontal and vertical energies of said gray scale image and cumulating both of the calculated energies;

providing a second energy image;

computing a maximum of both the energy images;

providing a maximum energy image;

binarizing said maximum energy image;

dilating said binarized image;

formulating different clusters based on the density of the dilated areas;

creating bounding boxes enclosing each cluster in the clustered image; and

forming an image of the document having image segments.

8. The method as claimed in claim 7, further comprising:

providing a database containing images of logos, product-marks and trademarks along with company and product information of a plurality of companies;

indexing said images of logos, product-marks and trademarks stored in said database;

receiving different image segments of a document;

matching said different image segments with said indexed images of logos, product-marks and trademarks stored in said logo database and providing matching image segments;

extracting said company and product information from said database corresponding to each of said matching image segments; and

appending non matching image segments containing new logos, product-marks or trademarks, together with company or product information, into the database.

9. The method as claimed in claim 7, wherein the step of pre-processing includes a step of noise removing.

10. The method as claimed in claim 7, wherein the step of clustering includes a step of using DBSCAN technique to compute clusters.