CONTENT RECOVERY AND IMAGE RETRIVAL IN IMAGE DATABASE CONTENT RETRIVING IN TEXT IMAGES

Scientific Journal Impact Factor (SJIF): 1.711
International Journal of Modern Trends in Engineering and
Research
www.ijmter.com
@IJMTER-2014, All rights Reserved 258
e-ISSN: 2349-9745
p-ISSN: 2393-8161
CONTENT RECOVERY AND IMAGE RETRIVAL IN IMAGE DATABASE
CONTENT RETRIVING IN TEXT IMAGES
K.Yuvasri[1]
, B.Arulkumar[2]
,S.Syed Farmhan[3]
1
Computer Science and Engineering, Vivekanandha College Of Engineering For Women
2
3
Abstract - Digital Images are used in magazines, blogs, website, television and more. Digital image processing
techniques are used for feature selection, pattern extraction classification and retrieval requirements. Color, texture
and shape features are used in the image processing. Digital images processing also supports computer graphics
and computer vision domains. Scene text recognition is performed with two schemes. They are character
recognizer and binary character classifier models. A character recognizer is trained to predict the category of a
character in an image patch. A binary character classifier is trained for each character class to predict the existence
of this category in an image patch. Scene text recognition is performed on detected text regions. Pixel-based layout
analysis method is adopted to extract text regions and segment text characters in images. Text character
segmentation is carried out with color uniformity and horizontal alignment of text characters. Discriminative
character descriptor is designed by combining several feature detectors and descriptors. Histogram of Oriented
Gradients (HOG) is used to identify the character descriptors. Character structure is modeled at each character
class by designing stroke configuration maps. The scene text extraction scheme is also supports for smart mobile
devices. Text recognition methods are used with text understanding and text retrieval applications. The text
recognition scheme is enhanced with content based image retrieval process. The system is integrated with
additional representative and discriminative features for text structure modeling process. The system is enhanced to
perform text and word level recognition using lexicon analysis. The training process is included with word
database update task.
Keywords- Text recognition, character recognition, pixel based layout,character descriptor,lexicon analysis.
I. INTRODUCTION
Content-based image retrieval (CBIR), also known as query by image content (QBIC) and content-based visual
information retrieval (CBVIR) is the application of computer vision to the image retrieval problem, that is, the
problem of searching for digital images in large databases. The term content in this context might refer colors,
shapes, textures, or any other information that can be derived from the image itself. Without the ability to examine
image content, searches must rely on metadata such as captions or keywords, which may be laborious or expensive
to produce. The term CBIR seems to have originated in 1992, when it was used by T. Kato to describe experiments
into automatic retrieval of images from a database, based on the colors and shapes present. Since then, the term has
been used to describe the process of retrieving desired images from a large collection on the basis of syntactical
image features. The techniques, tools and algorithms that are used originate from fields such as statistics, pattern
recognition, signal processing and computer vision.
There is growing interest in CBIR because of the limitations inherent in metadata-based systems, as well as the
large range of possible uses for efficient image retrieval. Textual information about images can be easily searched
using existing technology, but requires humans to personally describe every image in the database. This is
impractical for very large databases, or for images that are generated automatically, e.g. from surveillance cameras.
It is also possible to miss images that use different synonyms in their descriptions. Systems based on categorizing
images in semantic classes like cat as a subclass of animal avoid this problem but still face the same scaling issues.

International Journal of Modern Trends in Engineering and Research (IJMTER)
Volume 01, Issue 06, [December - 2014] e-ISSN: 2349-9745, p-ISSN: 2393-8161
CBIR systems can also make use of relevance feedback, where the user progressively refines the search results by
marking images in the results as relevant, not relevant, or neutral to the search query, then repeating the search
with the new information. Generally speaking, image content may include both visual and semantic content. Visual
content can be very general or domain specific. General visual content include color, texture, shape, spatial
relationship, etc. Domain specific visual content, like human faces, is application dependent and may involve
domain knowledge. Semantic content is obtained either by textual annotation or by complex inference procedures
based on visual content. This concentrates on general visual content descriptions.
II. LITERATURE SURVEY
2.1. Text Detection and Character Recognition in Scene Images with Unsupervised Feature Learning
Detection of text and identification of characters in scene images is a challenging visual recognition
problem. As in much of computer vision, the challenges posed by the complexity of these images have been
combated with handed signed features and models that incorporate various pieces of high-level prior knowledge. In
this paper, we produce results from a system that attempts to learn the necessary features directly from the data as
an alternative to using purpose-built, text-specific features or models. Among our results, we achieve performance
among the best known on the ICDAR 2003 character recognition dataset.Feature learning algorithms have enjoyed
a string of successes in other fields. To apply these algorithms to scene text applications, we will thus use a more
scalable feature learning system.
Specifically, we use a variant of K-means clustering to train a bank of features. Armed with this tool, we will
produce results showing the effect on recognition performance as we increase the number of learned features. Our
results will show that it’s possible to do quite well simply by learning many features from the data. Our approach
contrasts with much prior work in scene text applications, as none of the features used here have been explicitly
built for the application at hand. Indeed, the system follows closely the one proposed.
2.2. Top-down and Bottom-up Cues for Scene Text Recognition
The problem of understanding scenes semantically has been one of the challenging goals in computer vision for
many decades. It has gained considerable attention over the past few years, in particular, in the context of street
scenes. This problem has manifested itself in various forms, namely, object detection, object recognition and
segmentation Although these approaches interpret most of the scene successfully, regions containing text tend to
be ignored. One of the first things we notice in this scene is the sign board and the text it contains. Popular
recognition methods ignore the text and identify other objects such as car, person, tree, regions such as road, sky.
The importance of text in images is also highlighted in the experimental study conducted by Judd et al. They found
that viewers fixate on text when shown images containing text and other objects. This is further evidence that text
recognition forms a useful component of the scene understanding problem.Given the rapid growth of camera-based
applications readily available on mobile phones, understanding scene text is more important than ever. Although
character recognition forms an essential component of text understanding, extending this framework to recognize
words is not trivial. It consists of “roughly front parallel” pictures of signs are quite similar to those found in a
traditional OCR setting. In contrast, we show results on a more challenging street view dataset, where the words
vary in appearance significantly.
2.3. Real-Time Scene Text Localization and Recognition
Text localization and recognition in real-world scene images is an open problem which has been receiving
significant attention since it is a critical component in a number of computer vision applications like searching
images by their textual content, reading labels on businesses in map applications or assisting visually impaired.
Methods based on a sliding window limit the search to a subset of image rectangles. Methods in the second group
find individual characters by grouping pixels into regions using connected component analysis assuming that
pixels belonging to the same character have similar properties. Connected component methods differ in the

properties used. The advantage of the connected component methods is that their complexity typically does not
depend on the properties of the text and that they also provide a segmentation which can be exploited in the OCR
step.. The real-time performance is achieved by posing the character detection problem as an efficient sequential
selection from the set of Extremal Regions (ERs). In the first stage of the classification, the probability of each ER
being a character is estimated using novel features calculated with O(1) complexity and only ERs with locally
maximal probability are selected for the second stage, where the classification is improved using more
computationally expensive features.
A highly efficient exhaustive search with feedback loops is then applied to group ERs into words and select the
most probable character segmentation. It is further demonstrated that by inclusion of the gradient projection 94.8%
of characters are detected by the ER detector.
2.4 Detecting Texts of Arbitrary Orientations in Natural Images
The great successes of smart phones and large demands in content-based image search/ understanding have made
text detection a crucial task in human computer interaction. Although text detection has been studied extensively in
the past, the problem remains unsolved. The difficulties mainly come from two aspects: (1) the diversity of the
texts and (2) the complexity of the backgrounds.
On one hand, text is a high level concept but better defined than the generic objects; on the other hand, repeated
patterns and random clutters may be similar to texts and thus lead to potential false positives.Hence, combining the
strengths of specially designed features and discriminatively trained classifiers, our system is able to effectively
detect texts of arbitrary orientations but produce fewer false positives.To evaluate the effectiveness of our system,
we have conducted extensive experiments on both conventional and new image datasets. Compared with the state-
of-the-art text detection algorithms, our system performs competitively in the conventional setting of horizontal
texts.To evaluate the effectiveness of our system, we have conducted extensive experiments on both conventional
and new image datasets. Compared with the state-of-the-art text detection algorithms, our system performs
competitively in the conventional setting of horizontal texts. We have also tested our system on a very challenging
large dataset of 500 natural images containing texts of various orientations in complex backgrounds. On this
dataset, our system works significantly better than any of the existing systems, with an F-measure about 0.6, more
than twice that of the closest competitor
2.5.Scene Text Recognition using Part-based Tree-structured Character Detection
With the rapid growth of camera-based applications readily available on smart phones and portable devices,
understanding the pictures taken by these devices semantically has gained increasing attention from the computer
vision community in recent years. Most of the previous work on scene text recognition could be roughly classified
into two categories: traditional Optical Character Recognition (OCR) based and object recognition based. For
traditional OCR based methods, various binarization methods have been proposed to get the binary image which is
directly fed into the off-the-shelf OCR engine. Moreover, the loss of information during the binarization process is
almost unrecoverable, which means if the binarization result is poor, the chance of correctly recognizing the text is
quite small.
For scene character recognition, these methods directly extract features from original image and use various
classifiers to recognize the character. While for scene text recognition, since there are no binarization and
segmentation stages, most existing methods adopt multi-scale sliding window strategy to get the candidate
character detection results. To recognize the scene text, we build the CRF model on the potential character
locations. Character detection scores, spatial constraints and linguistic knowledge are used to define the unary and
pairwise cost function. The final word recognition result is acquired by minimizing the cost function.

III. PROBLEM DESCRIPTION
Camera-Based text information serves as effective tags or clues for many mobile applications associated with
media analysis, content retrieval, scene understanding and assistant navigation. In natural scene images and videos,
text characters and strings usually appear in nearby sign boards and hand-held objects and provide significant
knowledge of surrounding environment and objects. Text-based tags are much more applicable than barcode or
quick response code because the latter techniques contain limited information and require pre-installed marks.
To extract text information by mobile devices from natural scene, automatic and efficient scene text detection and
recognition algorithms are essential. Extracting scene text is a challenging task due to two main factors: 1)
cluttered backgrounds with noise and non-text outliers and 2) diverse text patterns such as character types, fonts
and sizes. The frequency of occurrence of text in natural scene is very low and a limited number of text characters
are embedded into complex non-text background outliers. Background textures, such as grid, window and brick,
even resemble text characters and strings. Although these challenging factors exist in face and car, many state-of-
the-art algorithms have demonstrated effectiveness on those applications, because face and car, have relatively
stable features. For example, a frontal face normally contains a mouth, a nose, two eyes and two brows as prior
knowledge. It is difficult to model the structure of text characters in scene images due to the lack of discriminative
pixel-level appearance and structure features from non-text background outliers. Further, text consists of different
words where each word may contain different characters in various fonts, styles and sizes, resulting in large intra-
variations of text patterns. To solve these challenging problems, scene text extraction is divided into two processes:
text detection and text recognition. Text detection is to localize image regions containing text characters and
strings. It aims to remove most non-text background outliers. Text recognition is to transform pixel-based text into
readable code. It aims to accurately distinguish different text characters and properly compose text words. This
paper will focus on text recognition method. It involves 62 identity categories of text characters, including 10
digits [0-9] and 26 English letters in upper case [A-Z] and lower case [a-z].
We propose effective algorithms of text recognition from detected text regions in scene image. In scene text
detection process, we apply the methods presented in our previous work. Pixel-based layout analysis is adopted to
extract text regions and segment text characters in images, based on color uniformity and horizontal alignment of
text characters. In text recognition process, we design two schemes of scene text recognition. The first one is
training a character recognizer to predict the category of a character in an image patch. The second one is training a
binary character classifier for each character class to predict the existence of this category in an image patch. The
two schemes are compatible with two promising applications related to scene text, which are text understanding
and text retrieval. Text understanding is to acquire text information from natural scene to understand surrounding
environment and objects.
Text retrieval is to verify whether a piece of text information exists in natural scene. These two applications can be
widely used in smart mobile device.The main contributions of this paper are associated with the proposed two
recognition schemes. Firstly, a character descriptor is proposed to extract representative and discriminative features
from character patches. It combines several feature detectors and Histogram of Oriented Gradients (HOG)
descriptors. Secondly, to generate a binary classifier for each character class in text retrieval, we propose a novel
stroke configuration from character boundary and skeleton to model character structure.

Fig. 1.1. The Flowchart Of Our Designed Scene Text Extraction Method
The proposed method combines scene text detection and scene text recognition algorithms. Similar to other
methods, our proposed feature representation is based on the stateof- the-art low-level feature descriptors and
coding/pooling schemes. Different from other methods, our method combines the low-level feature descriptors
with stroke configuration to model text character structure. Also, we present the respective concepts of text
understanding and text retrieval and evaluate our proposed character feature representation based on the two
schemes in our experiments. Besides, previous work rarely presents the mobile implementation of scene text
extraction and we transplant our method into an Android-based platform.
IV. PROPOSED SYSTEM
Scene text recognition process is performed to identify the text or string in a natural scene image. Text region
selection, Character descriptor and character structure analysis methods are used for text recognition process. The
system is enhanced to support text and word level recognition process. Content Based Image Retrieval (CBIR)
scheme is integrated with the system.
V. CONCLUSION
Scene text recognition process is performed to identify the text or string in a natural scene image. Text region
selection, Character descriptor and character structure analysis methods are used for text recognition process. The
system is enhanced to support text and word level recognition process. Content Based Image Retrieval (CBIR)
scheme is integrated with the system. The system improves the accuracy levels in the text recognition process.
Content based image search is supported by the system. Text and word level recognition scheme is used for the
scene understanding purpose. Text structure modeling is upgraded to improve the classification process.
REFERENCES
[1]A.Coates et al., “Text Detection And Character Recognition In Scene Images With Unsupervised Feature Learning,” in Proc. ICDAR,
Sep. 2011, pp. 440–445.
[2]A.Mishra, K. Alahari and C. V. Jawahar, “Top-Down And Bottom-Up Cues For Scene Text Recognition,” in Proc. IEEE Conf.
Comput. Vis. Pattern Recognit., Jun. 2012, pp. 1063–6919.
[3]A.Shahab, F. Shafait and A. Dengel, “ICDAR 2011 Robust Reading Competition Challenge 2: Reading Text In Scene Images,” in
Proc. Int. Conf. Document Anal. Recognit., Sep. 2011, pp. 1491–1496.
[4]C. Shi, C. Wang, B. Xiao and Z. Zhang, “Scene Text Recognition Using Part-Based Tree-Structured Character Detection,” in Proc.
CVPR, Jun. 2013.

[5]C. Yi and Y. Tian, “Localizing Text In Scene Images By Boundary Clustering, Stroke Segmentation, And String Fragment
Classification,” IEEE Trans. Image Process., vol. 21, no. 9, pp. 4256–4268, Sep. 2012.
[6]C. Yi, X. Yang and Y. Tian, “Feature Representations For Scene Text Character Recognition: A Comparative Study,” in Proc. 12th
ICDAR, 2013.
[7]C.Yao, X. Bai, W. Liu, Y. Ma and Z. Tu, “Detecting Texts Of Arbitrary Orientations In Natural Images,” in Proc. IEEE Conf. Comput.
Vis. Pattern Recognit., Jun. 2012, pp. 1083–1090.
[8]Chucai Yi and Yingli Tian, “Scene Text Recognition in Mobile Applications by Character Descriptor and Structure Configuration”
IEEE Transactions On Image Processing, Vol. 23, No. 7, July 2014
[9]D. L. Smith, J. Feild and E. Learned-Miller, “Enforcing Similarity Constraints With Integer Programming For Better Scene Text
Recognition,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., Jun. 2011, pp. 73–80.
[10]K. Wang, B. Bbenko and S. Belongie, “End-to-End Scene Text Recognition,” in Proc. Int. Conf. Comput. Vis., Nov. 2011, pp. 1457–
1464.
[11]L. Neumann and J. Matas, “Real-Time Scene Text Localization And Recognition,” in Proc. IEEE Conf. Comput. Vis. Pattern
Recognit., Jun. 2012, pp. 3538–3545.

CONTENT RECOVERY AND IMAGE RETRIVAL IN IMAGE DATABASE CONTENT RETRIVING IN TEXT IMAGES

CONTENT RECOVERY AND IMAGE RETRIVAL IN IMAGE DATABASE CONTENT RETRIVING IN TEXT IMAGES

More Related Content

What's hot (17)

Viewers also liked (9)

Similar to CONTENT RECOVERY AND IMAGE RETRIVAL IN IMAGE DATABASE CONTENT RETRIVING IN TEXT IMAGES (20)

More from Editor IJMTER (20)

Recently uploaded (20)

CONTENT RECOVERY AND IMAGE RETRIVAL IN IMAGE DATABASE CONTENT RETRIVING IN TEXT IMAGES