Pattern Analysis In Handwriting

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,

Example:

 The above figure shows that x and y are independent (or)dependent variables. we are taking 1-nearest neighbour .in this process only one nearest to the query point i.e. y4.so one nearest neighbour we can write y=y4.

 We can consider two nearest neighbours, in this case two points two points are nearer to ‘x’, which happen to be ‘y3’ and ‘y4’.so we are taking the average value

 k-nn measured as a distance function i.e. k=1 assigned to nearest neighbour distance function .

 The above distance functions are continuous variables instance of using hamming distance contain the numerical variables i.e. 0 and 1

When To Consider K-Nearest Neighbor:

i. When points are map to k^n. ii. When we are using K-NN less than 20 attributes per a particular data. iii. It’s containing heavy training data.

K-Nearest Neighbor Application:

 KNN applied optical character recognition its contains capital letters.

 Its have 14 attributes. Attributes means horizontal position of box, vertical position of box, width of box, height of box, total number of pixels in a box.

 It’s containing Validation: 10 fold cross validation.

 Its produces 26 classes

It builds a set of neurons identical to human brain and links the characteristics or features that are extracted from preprocessed document or characters. Each characters forms the input layer for neural network and the number of patterns( here the handwriting user) to be detected forms output layer. A neural network builds set of hidden layers and connects the input with output through these hidden layers. Hidden layers are perceived as truth tables that maps input with output with either eliminatory operations like logical OR, XOR on input data or advanced transform in high dimension feature space that defines a transfer function like laplace and z transform. Whichever transfer function and relationship a neural network uses to map the input features with output classes, it weights the features appropriately to distinguish the