Anirban-Chand / Image-Processing-And-Character-Segmentation

Image Processing and Character Segmentation for Bengali Script

Image-Processing-And-Character-Segmentation

Notes To Remember:

• PGM - Portable Graymap

• PBM - Portable Bitmap

• Magic Number - Magic numbers are the first few bytes of a file that are unique to a particular file type. These unique bits are referred to as magic numbers, also sometimes referred to as a file signature. These bytes can be used by the system to “differentiate between and recognize different files” without a file extension.

• Different Magic Numbers:

Magic Number	File Type	Extension	Type
P1	Portable Bitmap	PBM	ASCII
P2	Portable Graymap	PGM	ASCII
P3	Portable Pixmap	PPM	ASCII
P4	Portable Bitmap	PBM	Binary
P5	Portable Graymap	PGM	Binary
P6	Portable Pixmap	PPM	Binary

• Only P1, P2 magic numbers are used

• Noise Reduction:: (For noise reduction we use structuring element/kernel/window of different sizes.)

  • 1. Average/Mean Filter
  • 2. Median Filter

• Mean Filter - Superimpose the kernel on the image. Value at the center of the kernel is avg of the kernel.

  • Example: for 3x3 kernel - image[i][j] =(kernel[i-1][j-1]+kernel[i-1][j]+kernel[i-1][j+1]+kernel[i][j-1]+kernel[i][j]+kernel[i][j+1]+kernel[i+1][j-1]+kernel[i+1][j]+kernel[i+1][j+1])/9

• Median Filter - Superimpose the kernel on the image. Value at center of kernel is median of the kernel. We sort kernel elements & take median of them.

  • Example: image[i][j] = median(kernel[i-1][j-1], kernel[i-1][j], kernel[i-1][j+1], kernel[i][j-1], kernel[i][j], kernel[i][j+1], kernel[i+1][j-1], kernel[i+1][j], kernel[i+1][j+1])

• Binarization - Process of converting a multi-tone image into a two-tone image. Compute the threshold value of the multi-tone image using Otsu's Thresholding Algorithm. If value of a particular pixel is less than the threshold then we make it background(value = 1) and otherwise make that pixel foreground(value = 0).

• Otsu's Algorithm::

  • step-1: Histogram of the image
  • step-2: Compute Between-Class Variance
  • step-3: Find the Threshold value

• Threshold Value - Pixel value for which b/w class variance is maximum.

• Connected Components - 2 pixels are said to be connected if their pixel values are same and they are N4 or N8 neighbor of each other.

• Morphological Operations - Morphological operations are mathematical operations applied on binary images for analysis and processing.

  • Basic Types -
    • Dilation
    • Erosion
  • Composed Types -
    • Opening = erosion -> dilation
    • Closing = dilation -> erosion

• Line Detection - Detect where each line starts and ends.

• Horizontal Projection Profile - Sum of all pixels within a row.