Essay Example on Optical scanning is the first phase in OCR

Optical Scanning Optical scanning is the first phase in OCR As the first step a digital image of the original image will be captured The optical scanners which are used in OCR contains a transport mechanism and a sensing device The scanned image is usually a black print on a white background When OCR process occurs on the image it will convert multi level image onto bi level image This process in known as thresholding 24 Thresholding process is very important in OCR and also a fixed threshold is used there But in the practical scenario it is needed have advanced methods for thresholding Because it matters to achieve a good result Thresholding is performed on the scanner with the purpose to save memory space and computational effort 24 Preprocessing Segmentation Segmentation considered as the third phase in OCR Purpose of this phase is to define the elements of a pre processed image It is done be separating an individual image into sub images with individual characters in them In the OCR process it is needed to identify the areas where the texts are printed and also to differentiate them from figures and graphics as well Some problems can be occured in the segmentation phase Those problems can be divided into 4 categories 

They are Extraction of touching and fragmented characters 25 This kind of situation can be occurred with the dark photocopy and low threshold scanned documents In this situation joint characters are misunderstand as one single character and also a part of character can be misunderstood as a full symbol The joint characters can be segmented properly with the use of light photocopy or high threshold scanned document Differentiating text and noise Some documents images contain noise in those This noise is a main problem in character recognition process Sometimes the dots and accents in the documents images may be mistaken for noise and vice versa Misidentifying text for graphics or geometry and vice versa In the first scenario the text will not be sent to the recognition process And in the other case non text will be sent to the recognition process Segmentation of handwritten characters is more difficult scenario than segmentation of printed documents Because there is a huge variation in handwritten characters written by different people The major scenario that makes the segmentation process difficult is overlapping or touching characters Representation The fourth phase of OCR process is image representation An OCR system should be as simple as possible and also it should improve the accuracy of the recognition process

Therefore, it needs a more specific representation There are several image representation methods They can be categorized into 3 main groups They are Global transformation and series expansion Statistical representation Geometrical and topological representation Feature Extraction As the fifth phase in OCR feature extraction will be done Purpose of feature extraction is to remove redundancy from data It also means defining or measuring the exact shape of the basic character by capturing necessary characteristics features Feature extraction is needed when the input data is too large to process Feature extraction phase is very important and also it is considered as the most difficult phase of OCR process Before moving into feature extraction procedure there are 2 things to consider about they are feature extraction and feature selection Feature extraction is about which techniques to use to extract features from the image based on transformations or combination of original features Feature selection is about selecting the most suitable features from the input feature set to improve the accuracy of classification Accuracy of an OCR system can be increased by selecting an appropriate feature extraction technique There are several feature extraction techniques are available Feature extraction methods are based on pixel diagonal geometric etc Some of them are Structural approach Based on geometric and topological structures of characters Statistical approach Hybrid approach Linear discriminant analysis Principal component analysis Independent component analysis Chain code Scale invariant feature extraction Gradient based features Histogram Feature extraction techniques can be evaluated by using criteria mentioned below Robustness Robustness against Noise Robustness against Distortions Robustness against Style Font variation Robustness against Translation Robustness against Rotation Robustness against Size Practical use Speed of recognition Complexity of implementation Independence how independent the feature set is without requiring any supplementary techniques Training and Recognition