Aim : The Aim of the project is to develop a image processor whose output could be used by neural networks for text extraction from the input image.
Input : Image with single black text paragraph on white background as shown in sample
Output : Bounding rectange of each charactor in the input. But for representation purpose the output is an binary image with every charactor highlighted by enclosing it in a white rectangle.
Pre-requisites :
Modules - cv2, numpy and sys (you can download these using pip)
Python - Basic syntax, functions :str(), len(), range(), sum() (you can read about these functions here) and handling command line arguments in python.
OpenCV - imread(), dilate(), findContours(), contourArea(), Canny(), adaptiveThreshold(), erode(), bitwise_and(), minAreaRect(), getRotationMatrix2D(), warpAffine(), rectangle(), resize(), imwrite(), waitKey() and destroyAllWindows() (you can read about these functions here)
Numpy - ones(), where(), column_stack() (you can read about these functions here)
Brief algorithm:
1)Read the input image in Grayscale format.
2)Converted the image to binary format.
3)Cropped the image to minimize non-text region.
4)Removed any tilt from the text.
5)Determined the y-indices of the top and the bottom of each line of text (line segmentation).
6)In each line determined the x-indices of the beginning and the end of each charactor (charactor segmentation).
7)The bounding rectangle of each charactor is found by its x-indices and the y-indices of the line containing the charactor.
(for complete working of the project you can refer to docs)
Result - The bounding rectangle of a each charactor could be fed to a neural network to covert the charctor from optical to text format.
