University Of Texas At San Antonio, 6900 North Loop 1604 West, San Antonio, United States

Read Complete Research Material

UNIVERSITY OF TEXAS AT SAN ANTONIO, 6900 NORTH LOOP 1604 WEST, SAN ANTONIO, UNITED STATES

University of Texas at San Antonio, 6900 North Loop 1604 West, San Antonio, United States



Abstract

This paper examines a new technique for manipulating the curvature of the text and images applied to the segmentation of the dotted line. Some improvements are planned, including methods of treatment (a) the curvature in the detection and segmentation of the text (b) a determination of rotation angle (c) a technique to isolate only the pixels of the characters in the image (d) complete algorithm to stabilize the dot character (e segment characters) that can be printed too close or touching. An accuracy of 93% recognition rate is obtained by the application of new algorithms cons images of 90 bottles of Dasani water and Ozark, with different text and different lighting conditions. The filling algorithm exhibited better recognition of more than 20% compared to the characters not met.

Index Terms— character recognition, segmentation, rotation, angled text, connected characters, fill pixels image.

Khader Mohammad

University of Texas at San Antonio, 6900 North Loop 1604 West, San Antonio, United States

INTRODUCTION

The segmentation process is a crucial step in vision based recognition systems because it extracts meaningful regions for analysis ?[16]. The segmentation process attempts to decompose the text image into classifiable images called characters. There are many challenges in designing a single system that is capable of performing automatic recognition for curved text. In addition to the common segmentation issues, new challenges like curvature, character connectivity, varying text formats, and angled text all have a strong negative impact on the accuracy of the segmentation process and the results.

There is extensive literature on the problem of segmenting text areas in images ?[18]. Many methods are specifically designed to deal with special types of images, such as documents ?[19], WebPages [20], and pictures from newspaper ?[21]. Region based approach has been used in ?[4] to segment an angled text while ?[13] presents a new approach for segmentation. Most of the conventional techniques to segment characters deal with lines of solid text. It is very important to minimize the errors in this process so as to reduce the error-rate and increase successful recognition rate ?[3].

In ?[22] authors discussed text extraction from scene images. Researchers In ?[9] used object-based segmentation technique for character separation. The general challenges for text segmentation, as explained in ?[10], include image object, or characters that are too close, and images which are too blurred for any text to be recognizable.

The author in ?[1] presents a review of character segmentation. Most of the time, the type of the image drives the segmentation process. It is hard to define the optimal solution due to different constraints. A new methodology is proposed in ?[2] for character segmentation and recognition from grayscale images. In the proposed methodology, the character segmentation regions are determined by using projection profiles and topographic features extracted from the gray-scale images. They also presented a segmentation method for touching and overlapping characters; their algorithm ...
Related Ads