A Simple Classification of Binary Document into Vector Image or Scalar Image using Feed Forward Neural Networks with Back Propagation Training
G. Sudha · International Journal of Computer Applications · 2010
This paper describes a simple method to classify the binary document into scalar type or vector type.Any image can be basically classified into 1.Vector image 2. Scalar Image.Vector images are represented as display file containing set of drawing primitive commands (like move, line, curve, rectangle, circle etc) along with related parameters (like coordinates, angle, radius etc.).Scalar images are represented as pixel information ex.Bitmap.Thus memory required to store vector image is very less than of scalar image.Vector image is best to represent text since letters of text are generally formed by set of primitive drawing commands.If the vector image portions of a binary document are identified, then it can be represented as display file by the display file Generating-programs. Remaining portion of the file i.e. scalar portion can be compressed and kept/sent separately.Initially the binary document images are segmented using Constrained Run Length Algorithm (CRLA) which splits the entire binary image into labeled segment blocks.Next step is to calculate the features for each block according to the block size and data transition (zero-to-one / one-to-zero).Next, the blockfeatures are fed into input layer of Feed Forward Neural Network(FFNN) and it is trained using back propagation method to train the network to classify content of each block into scalar or vector type.Hidden layer and neurons in hidden layer can be developed using constructive approach.FFNN training can be improved using momentum, learning rate and bias if necessary.