Handwritten Character Recognition Using Image

Abstract

In the modern world, there is a huge increase in traffic congestion and the rate at which people are buying vehicles significantly rises every year. Traffic Signals are used to control the flow of traffic and are the most essential component of road safety and traffic control. The main objective of this paper is to propose a solution for controlling traffic signals automatically based on traffic density. This will be done by analyzing traffic density through an Infrared Sensor. A threshold value will be set above which traffic density will be considered ‘high’ and below which will be considered ‘low’.

A maximum value will also be set, above which time a signal cannot remain open. When a particular road has a high traffic density, it will have more time in the green signal compared to the roads that have a lower traffic density. Once the desired operation is chosen, it is sent directly to the traffic signal, which changes the timer and lights accordingly. The coding for the image processing is done in MATLAB and the main concepts used are Image Processing, Image Cropping and Conversion.

Introduction

The concept of text recognition has been around since the early 20th century. Text recognition is commonly known as OCR (Optical Character Recognition). Early Optical Character Recognition can be traced back to as early as 1914. These devices were mainly used to aid the blind.With the passing of time and the great advancements in the field of technology, these devices too have seen massive improvements of their own. The devices can now be used to translate printed text into various languages.

The system that has been proposed in this paper recognizes hand written text and converts it to a printed format that can then be viewed on screen. Despite the various advancements in the field of character recognition, studies involving the conversion of handwritten text have been quite rare. This is mainly because, unlike in the case of printed text conversion, handwriting varies from person to person and therefore, the software will have to identify and recognize each person’s handwritten character individually. Since the characters that have to be converted are handwritten texts, it is virtually impossible to create a database that contains the handwritten characters because, as mentioned before, handwriting varies from person to person.

The system that has been proposed in the paper uses the convex hull algorithm to identify and convert the handwritten text. This method is extremely effective as it greatly reduces the computation time and this can also recognize each person’s handwritten text individually. The proposed system uses Convex Hull Matching (CHM) to individually differentiate each letter. The paper also highlights the various steps used during the recognition and conversion of the handwritten text to a printed text format.

Existing System

Development and progress in various aspectsto the extraction of text information from an image have been around since the 20th century. These developments have been used for specific applications such as extraction from printed pages. Even though extensive research has been carried out, it is however not easy to design a series general-purpose system. That is because there many possible sources of variations and outcomes when extracting the text from the source. The Shaded contents from the textured background orfrom the low-contrast,complex images or images with variations in font size, style, color, orientation, and alignment. These variations make the problem very difficult which in turn makes drawing automatically very hard.

The commonly usedtext-detection methods can be classified into three categories. The first category consists of connected component-based methods, which assume that the text regions have uniform colors and satisfy certain size, shape, and spatial alignment constraints. These methods are not effective when the text have similar colours with background,so this would most likely result in improper detection.

The second category consists of texture based methods, which assume that the text regions have special texture. All these methods are less sensitive to background colors, so they may not be able todifferentiate between texts and the text-like backgrounds.

The third category consists of the edge-based methods. In this the text regions are detected under the assumption that the edge of the background and the object regions are sparser than those of the text regions. This kind of approaches is not very effective and suitable for detecting texts with large font size. When compared with theSupport Vector Machines (SVM) based method with the multilayer perceptions (MLP) based for textthe verification over four independent features which includes the distance map feature, grayscale spatial derivative feature, constant gradient variance feature and the DCT coefficients feature. Better results in detection are obtained by using SVM rather than by MLP. Multi-resolution-based text detection methods are often adopted to detect texts in different scales. Texts with different scales will have different So, in this paper we present and effective and alternative way for recognizing handwritten text.

Proposed System

The system which we have proposed in this paper is an advanced version of the existingSystem, with better text detection and recognition capabilities.The proposed structuralimprovements are:-Text Detection: This phase takes image input and decides whether it contains text or not. It also identifies the text regions in image using the convex hull method. Text Localization: Text localization merges the text regions to formulate the text contents and define the boundary around the text content.

Text Binarization: This step is used to segment the text contents from the background in the bounded text contents. By converting the given image to a grayscale image, then the binary value is determined. The output of text binarization is the binary image, where text pixels and background pixels appear in two different binary levels.Character Recognition: The final module of text extraction process is the character recognition. This module converts the binary text object into the convex hull image for which a value is determined.

The aim of Optical Character Recognition is to classify optical patterns for the handwritten text corresponding to alphanumeric or other characters. This done using the process of OCR which involves several step like segmentation, feature extraction, and classification. In principle, any standard OCR software can now be used to recognize the text in the segmented frames. A hard look at the properties of the candidate character regions in the segmented frames or image reveals that most OCR software packages will have significant or much difficulty to recognize the handwritten text. Documented images are different from natural images because they contain mainly text with a few graphics and images.

The determination of samples in the convex-hull of a set of high dimensions,is a time-complex task. To simplify this, a simple algorithm to compute an approximate convex hull is used. The algorithm used has the following advantages: it uses only one threshold which makes it is a deterministic algorithm, independent of the vertices initially considered; it also ensures that samples which correspond to maximum and minimum points for each dimension are present in the generated convex hull.A robust approach called convex hull matching (CHM) technique is used for registration of text contained images that differ from each other with Euclidean transformation. This method involves two steps, in the first step, point sets on the surface of the text image are extracted, and then the 3-D convex hull is constructed from the point sets and triangle patches on the surface of convex hulls are specified by predefining their normal vectors.

In the second step, each edge of the referenced triangle is compared with all the edges of the triangle in other point set to find the equal pair set and also to obtain the scaling factor. The transformation parameters of each triangle pairs including rotation and translation are optimized by minimizing the Euclidian distance between the corresponding vertex pair. Using these values the handwritten text is converted to the machine format text.

Algorithm

1. Start
2. Scan the handwritten text
3. Convert the color image into grayscale
4. Do segmentation, separatinglines from textual image
5. Apply Convex Hull to the scanned text
6. Compare the results to the closest match in the database
7. Load the matched data into the database.

The process of detection and recognition of hand written text is a series of orderly steps:

The hand written text is detected through an image input device (camera/video camera). This verifies if a legible text is present or not. If the text is not clear then the process is brought to a halt.

The colored image is converted into its corresponding grey image which makes it easier to be processed.

The recorded image of the text is then segregated, letters from blank spaces. The blanks spaces are removed and only the text part is sent further for recognition.

Convex hull algorithm is applied to the segmented result. Dimensional values are generated and sent for classification.

The alphanumerical characters are classified in their domain according to the values obtained in the convex hull procedure. CONVEX HULLConvex hull is a geometrical problem that can be solved computationally. The algorithm is about circumscribing a set of points by a polygonal figure called convex polygon. In other words convex polygon is a polygon that can house all the points in the given plane.Convexity of a polygon is measurable trait that is amenable to the analysis of its shape. In the text recognition procedure all the letters are associated with certain points that act as guidance points, they then form convex polygon. The database consists of separate unique values corresponding to each alphanumeric. The convex polygon is processed and a ratio value is obtained corresponding to its dimensions, these dimension ratios are compared to the values that are stored in the database and the closest match is picked and displayed.

Conclusion

The process of image processing makes it possible to detect hand written text. Now hand written documents can be scanned and made into computer typed document. This in the long run is of huge importance as people now don’t have to type the entire text to upload it into the cloud.The concept of image processing has a lot of merits and could have a lot of future applications