Cv2 finding pictures in another picture

Question

A big picture (300*300 pixels) divided into 3 areas. 2 small ones (AAA.png and BBB.png) are portions of the big picture.

3-areas.png

AAA.png

BBB.png

I want to know in which areas they located in, i.e. find small pictures in the big picture. The ideal output would be: "AAA.png is in the left"; "BBB.png is in the right".

I have these codes running:

import cv2, os
import numpy as np

big_pic = cv2.imread("c:\\TEM\\3-areas.png")

left_area = big_pic[0:300, 0:100]   # [y,y1 : x,x1]
mid_area = big_pic[0:300, 100:200]
right_area = big_pic[0:300, 200:300]

AAA = cv2.imread('C:\\TEM\\AAA.png')
AAA_res = cv2.matchTemplate(left_area,AAA,cv2.TM_CCOEFF_NORMED)

threshold = 0.99
AAA_loc = np.where(AAA_res >= threshold)
a_x_cor = list(AAA_loc[0])
a_y_cor = list(AAA_loc[1])
print a_x_cor, a_y_cor

BBB = cv2.imread('C:\\TEM\\BBB.png')
BBB_res = cv2.matchTemplate(right_area,BBB,cv2.TM_CCOEFF_NORMED)

BBB_loc = np.where(BBB_res >= threshold)
b_x_cor = list(BBB_loc[0])
b_y_cor = list(BBB_loc[1])

print b_x_cor, b_y_cor

I want to simplify it by making for loops. What I tried:

big_pic = cv2.imread("c:\\TEM\\3-areas.png")

list_of_areas = {
left : [0,300, 0,100],
mid : [0,300, 100,200],
right : [0,300, 200,300]}

small_picture_folder = "C:\\TEM\\"
small_pictures = []

root, dirs, files = os.walk(small_picture_folder).next()

for path, subdirs, files in os.walk(root):
    for f in files:
        if ""AAA"" in f or ""BBB"" in f:
           small_pictures.append(small_picture_folder + f)

for key, value in list_of_areas.iteritems():
    for y,y1,x,x1 in value:
        target_area = big_pic[y:y1, x:x1]

The "for y,y1,x,x1 in value:" line gives error: "TypeError: 'int' object is not iterable."

What would be the best way to achieve it? Thank you.

Answer 1

In your outer loop, iterating the dictionary entries

for key, value in list_of_areas.iteritems():

the value is a list of four integers, e.g. [0, 300, 0, 100]. So, when you now try to iterate this list using

for (...) in value:

you'll get four iterations, and in each iteration, exactly one list item will be "returned" at a time. But now, you try to assign this "returned", single integer value to four variables simultaneously by

for y, y1, x, x1 in value:

hence the error! So, from my point of view, the easiest solution would be, to skip the inner loop, and directly assign all items in value to your four variables, like so:

for key, value in list_of_areas.iteritems():
    y, y1, x, x1 = value
    target_area = big_pic[y:y1, x:x1]

Hope that helps!

Answer 2

what you tried to do is an objet detection algorithm. take a look at this opencv tutorial. https://docs.opencv.org/2.4/doc/tutorials/features2d/feature_homography/feature_homography.html to improve speed, try to replace Surf keypoint by orb keypoint (for your use case it will be probably good enougth, (if you do that, do not forget to modify the norm from L2 to Hamming )) it is a generalization of your problem. kind regards

Answer 3

This is how to detect multiple templates and crop ROI

import cv2
import numpy as np
templates = []
templ_shapes = []
threshold = 0.9
# get 1st template
temp1 =cv2.imread("YourImagePath1.png",cv2.IMREAD_GRAYSCALE)
templates.append(temp1)
templ_shapes.append(temp1.shape[:: -1])

# get 2nd template
temp2 =cv2.imread("YourImagePath2.png",cv2.IMREAD_GRAYSCALE)
templates.append(temp2)
templ_shapes.append(temp2.shape[:: -1])

# templ_shapes contains (width and height) dimensions of your templates
# so use them to display bounding box

# detect and get ROI
target =cv2.imread("YourImage_Target_Path.png",cv2.IMREAD_GRAYSCALE)
count = 0
for template in templates:
result = cv2.matchTemplate(target, template, cv2.TM_CCOEFF_NORMED)
loc = np.where(result >= threshold)
for pt in zip(*loc[::-1]):
    # pt is the top left corner of the template location so pt[0] + templ_shapes[count][0] is the X bottom right ...
    # templ_shapes[count][0] is the width and templ_shapes[count][1] is height
    cv2.rectangle(target, pt, (pt[0] + templ_shapes[count][0], pt[1] + templ_shapes[count][1]), (0, 0, 255), 2)
    roi = target[pt[1]:pt[1] + templ_shapes[count][1] ,pt[0] :pt[0] +templ_shapes[count][0]] 
count+=1