Get array from confusion matrix image. And save it to xlsx file for manual correction.

conf_matrix_ocr.py

from __future__ import print_function
import cv2
import pytesseract
import numpy as np
from os import listdir
from os.path import join
import pandas
import xlsxwriter


def getCFMatrix(imgpath, class_num=4, start_position=(40, 128), block_height=95, block_width=93):
    '''
    Convert confusion matrix to numpy matrix via ocr.
    :param imgpath: Confusion matrix file path.
    :return: Numpy matrix.
    '''
    img = cv2.imread(imgpath)
    ErrorRows = []
    CFMatrix = np.zeros([class_num, class_num])
    for i in range(class_num):
        for j in range(class_num):
            cropped_img = img[start_position[0] + block_height * i:block_height * (i + 1) + start_position[0],
                          start_position[1] + block_width * j:start_position[1] + block_width * (j + 1)]
            text = pytesseract.image_to_string(cropped_img)
            CFMatrix[i, j] = float(text)
    RowSum = CFMatrix.sum(axis=1)
    for i in range(RowSum.shape[0]):
        if np.abs(RowSum[i] - 1.0) > 0.00005:
            ErrorRows.append(i)
            print('-'*20)
            print("Error occurs in row {0}".format(i+1))
            print(imgpath)
    return CFMatrix, ErrorRows

def func_for_sort_eatract_int_from_file_name(name):
    return int(name.split('.')[0])


def handleDFentry(frame):
    assert frame.shape[0] == 1
    assert frame.shape[1] == 9
    code_location = frame['code_location'].values[0]
    model_location = frame['model_location'].values[0]
    if pandas.isnull(model_location):
        print("Nan value",frame)
        return np.zeros([4, 4]), None ,[0,1,2,3]
    log_location = join(code_location, 'logs', model_location[model_location.find('/') + 1:], 'confusion_matrix')
    cfmLists = list(listdir(log_location))
    cfmLists.sort(key=func_for_sort_eatract_int_from_file_name)
    cfm_file_path = join(log_location, cfmLists[-1])
    CFMatrix, ErrIndex = getCFMatrix(cfm_file_path)
    return CFMatrix, cfm_file_path, ErrIndex


if __name__ == '__main__':
    df = pandas.read_excel('./Experiments.xlsx')
    # Rename dafaframs header.
    df.columns = ['version', 'baseline', 'name', 'description', 'accuracy', 'commit', 'code_location', 'model_location',
                  'device']
    # frame = df[df['version'] == 'v51']
    # Create an new Excel file and add a worksheet.
    workbook = xlsxwriter.Workbook('Results.xlsx')
    rows = df.shape[0]
    for row in range(rows):
        frame = df[row:row+1]
        CFMatrix,cfm_file_path, ErrIndex = handleDFentry(frame)
        error_format = workbook.add_format({'bold': True, 'font_color': 'red'})
        worksheet = workbook.add_worksheet(frame['version'].values[0])
        for i in range(4):
            for j in range(4):
                if i in ErrIndex:
                    worksheet.write(i+1,j+1,CFMatrix[i,j],error_format)
                else:
                    worksheet.write(i+1, j+1, CFMatrix[i, j])
            worksheet.write_formula(i+1,5,'=SUM(B{0}:E{0})'.format(i+2), value=np.sum(CFMatrix[i,:]))
        if cfm_file_path:
            worksheet.insert_image('G2', cfm_file_path,{'x_scale': 1.0, 'y_scale': 1.0})
        worksheet.write('A1','GT/Pred')
        worksheet.write('A2','Sit')
        worksheet.write('A3','Walk')
        worksheet.write('A4','Ride')
        worksheet.write('A5','Stand')
        worksheet.write('B1','Sit')
        worksheet.write('C1','Walk')
        worksheet.write('D1','Ride')
        worksheet.write('E1','Stand')
        worksheet.write('A7','meanAcc')
        worksheet.write('F1','Sum')
        worksheet.write_formula(6,1,'=AVERAGE(B2,C3,D4,E5)',value=np.trace(CFMatrix)/4)
        worksheet.set_column('A:E', 10)
    workbook.close()