KLKS_test.py

import numpy as np
import pandas as pd
import cv2
import os,collections
from scipy import fftpack
from sklearn.metrics.pairwise import cosine_distances as edist
import pickle

model=open('model','rb')
ref_size,Reference_dist,dct_Reference=pickle.load(model)

Test_df=pd.read_csv('TestFile.data',names=['Test','true'],sep='\t')
Test=list(Test_df['Test'])
true=list(Test_df['true'])
test_size=(len(Test))
dct_Test=[]

for image in Test:
    name='./Images/'+image
    img=cv2.imread(name,0)
    row,col= img.shape[:2]
    ratio=float(row)/float(col)
    if col>row:
        dim=(512,int(ratio*512))
    else:
        ratio=1/ratio
        dim=(int(ratio*512),512)
    im2=cv2.resize(img,dim,interpolation=cv2.INTER_AREA)
    
    ret,binarised = cv2.threshold(im2,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU)
    if col>row:
        resized=cv2.copyMakeBorder(binarised,top=0,bottom=512-binarised.shape[0],left=0,right=0,borderType= cv2.BORDER_CONSTANT, value=255)
    else:
        resized=cv2.copyMakeBorder(binarised,top=0,bottom=0,left=0,right=512-binarised.shape[1],borderType= cv2.BORDER_CONSTANT, value=255)
    
    kernel = np.ones((5,5),np.uint8)
    resized = cv2.erode(resized,kernel,iterations = 1)
    
    resized=np.array(resized,dtype=np.float)
    
    dct=fftpack.dct(fftpack.dct(resized.T, norm='ortho').T, norm='ortho')
    dct_copy=dct[:20,:20]
    dct_Test.append(dct_copy.reshape(400))

Test_dist=edist(dct_Test,dct_Reference)

pickle.dump([test_size,Test_dist,true],open('result','wb'))