Letters Extraction in Sign Board Using Various Optimization Techniques

Letters Extraction in Sign Board Using Various Optimization Techniques

Priti1 , Assistant Professor, CSE Department,

GEC, Panipat1

Gyanender Kumar2 Assistant Professor, ECE Department, GEC, Panipat2

Abstract- Accurate localization of willy-nilly deployed sensor nodes is critically important in wireless sensor networks (WSNs) deployed for observation and trailing applications. The localization challenge has been displayed as a multidimensional world optimization problem in earlier literature. Various swarm intelligence algorithms are projected for correct localization. The untapped enormous potential of the BFO algorithm has inspired the research presented in this paper. The ABC algorithm has been investigated as a tool for anchor-assisted sensor localization in WSNs. Results of Matlab simulation of BFO based various stage localization has been given. Further, the results are compared with those of the localization methodology supported the particle swarm optimization (PSO) algorithm. A comparison of the performances of BFO and PSO algorithms has been given in terms of the no. of nodes localized, localization accuracy and the computation time. The results show that the ABC technique delivers higher accuracy of localization than the PSO technique does; but, it takes longer to converge. This leads to exchange between speed and accuracy of localization in WSNs.

Key Words:-Text extraction, Text Localization, text recognition, PSO,BFO.

1. INTRODUCTION

   The text portion embedded in born-digital pictures, prevalent on the Web, carry salient semantic information such as advertisements and security-related information. Therefore, extracting text information from born-digital images enhances the semantic relevance of web content for indexing and retrieval. Usually, a text information extraction system consists of three steps: text localization, text segmentation & text recognition. Text localization is critical to the overall system performance and is suffering from variable image background, text color and layout.

   Many methods have been proposed for text localization in images, and they roughly fall into two categories: connected component (CC)-based and texture-based.CC- based methods have achieved state-of-the-art results on various public datasets (Kai Chen, 2016). Extracting candidate text CCs is the most critical step for those methods. On one hand, as many text CCs as possible should be recalled. On the other hand, fewer non-text CCs should be generated so that text CCs are more easily identified and grouped into text lines. Researchers frequently use color, stroke width transform (SWT) and maximally stable extremal regions (MSERs) to cluster pixels into CCs. Color clustering based methods

   (Yuanyuan Feng , 2015 ; Kuntpong Woraratpanya , 2014) adopt strategies such as k-means to segment an image in the hope that pixels belonging to the same text CCs are segmented into the same sub-image. Texture based methods are based on the observation that text regions in images have different textural properties in contrast to non- text regions. Some textual based methods ( J. Canny,1986 ;

   H. Chen et.al , 2011) slide a sub-window in multi-scales through all locations of the image using a trained classifier to decide whether the sub window contains text or not. The exhaustive search makes the computation costly. There are also other methods A. (Clavelli et.al, 2010; T. E. de Campos et. al, 2009) which generate text region proposals first and decide whether each region proposal contains text or not. Designing effective features for discriminating textual regions from non-textual regions is the main difficulty of textual based methods.

   The proposed method is CC-based. Unlike that many methods have been proposed for scene text detection, few works have been published specifically for born-digital images. Because born-digital images present different characteristics from scene images, it is not necessarily true that methods developed for scene images are appropriate for born-digital images. Text strokes in born-digital images usually have complete contours and pixels on the contours have high divergence compared with the adjacent non-text pixels. This is often not true for text in scene images due to non-ideal camera-capturing environment.

   The rest of the paper is organized as follows. Section II outlines literature survey. Proposed Algorithm to analyzed F- Score is discussed in Section III. The Performance parameter is analyzed in Section IV. Section V is concentrated on the simulated result of Viterbi Decoder. The conclusions are given in Section VI.

2. LITERATURE SURVEY

   Kai Chen proposesa new CC based method for text localization in born-digital images. The proposed method generates character candidates effectively by first detecting text contours and stroke interior regions separately and then combining them. The CCs undergo CC filtering, line grouping and line classification to give the final result. Their method has achieved state-of-the-art performance on the born-digital dataset of ICDAR2013 Competition, convincingly demonstrating the effectiveness.

   Yuanyuan Fengpresents a text detection method based on Extremal Regions (ERs) and Corner-HOG feature. Local Histogram of Oriented Gradient extracted through corners is used to effectively prune the non-text components. Experimental results show that the Corner Histogram oriented Gradient based pruning method can discard an average of 73.06% of all ERs in an image while preserving a recall of 80.51% of the text components.

   Kuntpong Woraratpanya tells Thai Text localization and extraction in natural scenes is still a major challenge in current applications. However, the efficiency of recognition rates depends on text localization, i.e., the higher purity of text-background decomposition leads to the higher accuracy rate of character recognition. In order to achieve this purpose, the text-background decomposition methods, namely adaptive boundary clustering (ABC) and n-point boundary clustering (n- PBC), are proposed to improve a precision of text localization.

   Khalid Iqbal tells about text localization

   .Text localization in scene images is an essential task to analyze the digital image contents. In this work, a Bayesian network scores using K2 algorithm in conjunction with the features based effective text localization method with the help of maximally stable extremal regions (MSERs).

   Lukas Neumann gives an end-to-end real- time text localization and recognition method is presented. Its performance is achieved by posing the character detection and localization problem as an efficient sequential selection from the set of Extremal Regions. Mohammad ShorifUddin, Madeena Sultana, Tanzila Rahman, and Umme Saym 2012 their main objective is to extract text from images.In this method, author discussed a approach for detecting and localization text from scene images based on morphological features. Many researchers have been working on the development of techniques to extract texts from a scene images. Nirmala Shivananda and Naga bhushan proposed a hybrid method for separating text from color document images. But this technique can't extract text from difficult graphics. Partha Pratim Roy, JosepLlad' os and Umapada Pal proposed a method for differentiate text from color map based on CC analysis and grouping of characters in a string. These methods can detect the characters connected to graphics and separate them. But some of the characters can't be differentiate through connected component analysis. The algorithm of Antani and R.Kasturi works well for text separation from mixed text/graphics image, but it makes an assumption which is not practical that character components in texts are aligned straight and does not touch

   or overlap with graphics. For improving accuracy they used modified morphological filter and also proposed a clustering methods. Moreover, their method reduces noise in the resultant image. Experimental result confirms the dominance of their approach. With this method F-Score of 92.8% is achieved. RM. Chandrasekaran, P. Natarajan implemented an approach in 2012. In this technique, they have presented a approach for text localization and recognition in images. First, to remove any noise in the input image, the Median filter is used. Edges are detected using ZCED algorithm. Then the morphological dilation operation is applied for object localization.

3. PROPOSED ALGORITHM

   For Text localization, Particle Swarm optimization & Bacteria Forging optimization are used. These briefly explained as

   1. PSO Technique

      Particle Swarm Optimization modeled after the social behavior of birds in a flock. PSO is a population based search process where individuals act as particles and grouped act as a swarm. A candidate solution to the optimization problem is represented by swarm particle. In a PSO system, each particle is flown through the multidirectional search space&according to its own experience adjusting its position in search space and that of neighborhood particles. Therefore, the best position encountered by particlemake a particular use & their neighbors to position itself toward an optimizesolution. The effect is that still searching a large area around the best solution while particles fly toward a minimum. Using predefined fitness functions, the performance of each particle is measured which condense the characteristics of the optimization problem.

   2. BFO

   In order get maximum energy obtained per unit time, Bacteria search for nutrients.In this technique, communication between two bacterium takes place by sending the signals. A bacterium takes foraging decisions after considering previous factors. The process, in which a bacterium searching for nutrients and moves by taking very small steps, is called chemot axis and key idea of BFOA is repeating movement of imaginary bacteria in the problem search space. Since, BFOA has drawn the attention of researchers from diverse fields of knowledge especially due to its biological initiative and elegant structure. Different researchers are trying to hybridize BFOA with different other techniques in order to explore its global and local finding properties separately. This technique has already been applied to many problems and proved its power over many variants of Generic Algorithm and PSO. During foraging of the real bacteria, shifting is achieved by a set of tensile flagella. The two basic operations performed by a bacterium are E coli bacterium to tumble or swim.

4. PERFORMANCE PARAMETER It can be done on the basis of following parameters.

   1. False Positives (FP) / False alarms are those regions in the image which are actually a text, but have been detected by the algorithm as text.

   2. False Negatives (FN)/ these are those regions in the image which are actually texts, but have not been detected by the algorithm.

   3. Precision rate (p) is defined as the average ratio of correctly detected characters text without image to the sum of correctly detected characters text plus false positives as represented in equation. p = Correctly detected characters/ [Correctly detected characters +FP]

   4. RRC (Recall rate) = (No. of extracted characters text in image / No. of characters text in image) x 100.

   5. F-score is the H.M of the precision rate and recall rates

5. RESULT & DISCUSSION

   The Final extracted images are as shown in below fig. The comparison result is given as shown in Table 1.

   (a)

   (b)

   (c)

   (d)

   Fig 1 Final Text Extraction from Different Images.

   Fig 1gives final detected text from the image. These texts counts & find the recall rate & precision rate. Hence find F- score.

   As more is the Fscore, algorithm is more efficient.

   In our research, we have taken 10 images for text detection. On the basis of PSO & BFO algorithms recall rate and precision rate can be calculated. Table 1 gives Comparison of F-Score of the different images. The 1st Image has more FScore as compare to others.

   Table 1. Comparison of F- Score of different Images

   Sr. No.	Precision Rate	Recall Rate	F score
   1.	86	79	82.35
   2.	85	74	79.124
   3.	71	59	64.45
   4.	65	50	56.89
   5.	62	45	52.44
   6.	54	29	38.31
   7.	48	26	33.63
   8.	36	23	25.70
   9.	35	21	26.18

6. CONCLUSION

A method of character text string extraction from scene image is discussed and implemented. It cant only detect single and multiple texts but also detect arbitrary oriented text. It is based on mathematical morphology and OCR technique which can deal with various cases of scene images. Because in mathematical morphology, top-hats transformation (TT) provides an brilliant tools for extracting bright or dark objects from complex background. But for many complicated extraction problems, TT alone cant derive better result. For this reason, we discuss this method which uses segmentation of images using mathematical morphology and extraction of text using OCR. The experimental results appear encouraging to demonstrate the efficiency of PSO and BFO for shape analyzing and detecting. We intended to proceed with our study on text extraction from image to improve its

accuracy. With this approach F-score of 83.35 is achieved. Finally a character recognition system on scene image using PSO & BFO is suggested to be constructed.

REFERENCES

1. Kai Chen (2016), Effective Candidate Component Extraction for Text Localization in Born-Digital Images by Combining Text Contours and Stroke Interior Regions, 12th IAPR Workshop on Document Analysis Systems (DAS),pp 352 357

2. Yuanyuan Feng (2015) Scene text localization using extremal regions and Corner-HOG feature IEEE International Conference on Robotics and Biomimetics (ROBIO)pp881 886.

3. Kuntpong Woraratpanya (2014) , Text-background decomposition for thai text localization and recognition in natural scenes , 6th International Conference on Information Technology and Electrical Engineering (ICITEE),pp1-6

4. Khalid Iqbal (2014), Bayesian network scores based text localization in scene images,International Joint Conference on Neural Networks (IJCNN), 2218 2225.

5. Lukas Neumann (2015), Real-Time Lexicon-Free Scene Text Localization and Recognition, IEEE Transactions on Pattern Analysis and Machine Intelligence , Vol. 38 , Issue 9 pp 1872 1885.

6. MohammadShrifUddin,MadeenaSultana,TanzilaRahman,and UmmeSaymaBusra(2012),Extractionoftexts from a scene image using morphology based approach, IEEE Transactions on image processing, Vol.10, pp.306-309.

7. ] N i r m a l a Shivan and aand P. Nagabhushan (2009), Separation of Foreground Text from Complex Background in Color Document Images, IEEE Transactions on Image Processing, vol.10, pp.306-309.

8. Partha Pratim oy, Josep L lad´os and Umapada Pal (2007),Text / Graphics Separation in Color Maps ,IEEE Transactions on Image Processing, vol .7, pp.545-551.

[ 9 ] D. Crandall, S. Antani, and R. Kasturi (2001), Robust Detection of Stylized Text Events in Digital Video, Proceedings of International Conference on Document Analysis and Recognition, pp. 865-869.

1. R. Chandrasekaran, R. M. Chandrasekaran, P. Natarajan (2012) ,Text localization and extraction in images using mathematical morphology and SVM,IEEE International Conference(ICAESM) , pp 265-269

2. J. Canny (1986) A Computational Approach to Edge Detection. IEEE Trans. PAMI, 8(6):679698.

3. H. Chen, S. S. Tsai, G. Schroth, D. M. Chen, R. Grzeszczuk, and B. Girod (2011) ,Robust text detection in natural images with edge-enhanced maximally stable extremal regions , In Proc. ICIP, pages 26092612

4. Clavelli, D. Karatzas, and J. Llados (2010) , A Framework for the Assessment of Text extraction Algorithms on Complex Colour Images , In Proc. 9th DAS, pages 1928.

5. T. E. de Campos et.al (2009) Character recognition in natural images , In Proc. VISAPP, pp 220-226

6. R. O. Duda, P. E. Hart, and D. G. Stork (2006) .Pattern Classification. Wiley, 2 edition ,pp 434-439

7. B. Epshtein, E. Ofek, and Y.Wexler (2010). Detecting text in natural scenes with stroke width transform. In Proc. 23rd CVPR, pages 29632970.

8. B. Gatos, K. Ntirogiannis, and I. Pratikakis (2009), Document Image BinarizationContest ,In Proc. 10th ICDAR, pages 1375-1382.

9. GNU Image Manipulation Program (GIMP). http://www.gimp.org/.

10. R. C. Gonzalez and R. E. Woods (2002) ,Digital Image Processing. Prentice Hall, 2 edition.