Handwriting Analysis: A General Introduction To Handwriting Recognition

CHAPTER-1

INTRODUCTION

1.1 General Introduction to Handwriting Recognition Every individual have unique handwritings. Many person’s handwritings may have similar characteristics and styles in common, acquired when these they learned to write. They tend to take on individual styles with person age and their handwriting will show additional changes. Hence handwriting analysis is one key aspect in person identification.

In pattern recognition terminology Handwriting is perceived as identifying a person particular style of writing and skill. As there are no quantitative measures for such “styles” and writing styles cannot be represented with any physical world entities other than common observations like “cursive”, “bold” and so on,

Like most of the pattern recognition problems, handwriting recognition system can also be presented as a generic system with its functionalities being modularized and well described. Once the parts and methods adopted currently in this area is well understood, problem definition and proposed methodology gets lot easier to understand.

Though handwriting recognition system is as old as the history of written text is extended, computer based handwriting detection has gained momentum from late 80’s and early 90’s. With personal computers having the powers unimaginable even a decade back, the essence of this area has shifted from trying to match a handwriting style to recognize the writer of a handwritten document.

As such the challenge in hand writing recognition system is increasing the accuracy of detection and classification. For example zip code contains the 0-9 characters. If the authority has got access to only a letter with printed address and only the zip code is written by the user, the system should still be able to identify the

The result of binary filtered document is shown in f

Fig4.2 Filtering and Binarization of User 1 Sample Once the document was binary, pixels with ones, i.e. the written part could be conceived as a group of pixels being linked through the neighbourhood context. As such a bounding box can be obtained across all bounded regions which essentially defines the written words. Bounding boxes are the rectangles which defines the diagonal coordinates of the bounded region. Words can be segmented through these bounded coordinates

.

Fig 4.3 Bounding box across all the words In training process statistical features from all the images present in dataset folders are extracted. Each user’s words are segmented in the folder with names “1”,”2” for first, second writer respectively. Therefore in training process, the system scans every word stored in 1 directory while training user1’s handwriting