first commit

README.md

@@ -0,0 +1,17 @@
+# Image recognition using the Hopfield Network
+
+_Image recognition of The Simpsons using the Hopfield network. An implementation with Python._
+
+## Built with 🛠️
+
+* [OpenCV-python](https://github.com/opencv/opencv-python)
+* [Numpy](https://numpy.org/)
+* [Neurolab](https://pypi.org/project/neurolab/)
+* [Matplotlib](https://matplotlib.org/)
+
+## Screenshot 📖
+
+![Screenshot](image.PNG)
+
+---
+⌨️ with ❤️ by [hrypasato](https://github.com/hrypasato) 😊

hopfield.py

@@ -0,0 +1,66 @@
+"""
+Example of use Hopfield Recurrent network
+=========================================
+
+Task: Recognition of Simpsons
+
+"""
+import cv2, glob
+import numpy as np
+import neurolab as nl
+import matplotlib.pyplot as plt
+
+IMAGE_SIZE = (48, 48, 3) #size 48x48 pixels
+TEST_PATH = './data/test/*.png'
+TRAIN_PATH = './data/train/*.png'
+
+def img2array(name):
+  img = cv2.imread(name) #read binary image
+  img = img.flatten() #Return a copy of the array collapsed into one dimension.
+  img[img == 255] = 1
+  return img
+
+def array2img(array):
+  array[array == -1] = 0
+  array *= 255 # assing white spaces
+  img = np.reshape(array,IMAGE_SIZE) #transform one dimension array to multidimension array
+  return img
+
+def array2float(array):
+  tmp = np.asfarray(array)
+  tmp[tmp == 0] = -1 
+  return tmp
+
+def show_images(images):
+  fig = plt.figure()
+
+  ax = fig.add_subplot(1, 2, 1)
+  imgplot = plt.imshow(images[0])
+  ax.set_title('Test')
+
+  ax = fig.add_subplot(1, 2, 2)
+  imgplot = plt.imshow(images[1])
+  ax.set_title('Result')
+
+  plt.show()
+
+
+
+target = []
+for file in glob.glob(TRAIN_PATH):
+  array = img2array(file)
+  target.append(array)
+
+target = array2float(target)
+
+net = nl.net.newhop(target) # Create and train network
+
+
+img_test = img2array('./data/test/lisa1.png')
+test = array2float(img_test)
+
+out = net.sim([test]) #test network
+
+out_image = array2img(out[0]) #output network image
+img_test = array2img(test) #test image
+show_images([img_test, out_image])

utils/data_image.py

@@ -0,0 +1,23 @@
+#extract data from image and save into excel
+#Getting Pixel Values
+from skimage import io
+import pandas as pd
+import glob
+
+TEST_PATH = '../data/test/*.png'
+TRAIN_PATH = '../data/train/*.png'
+TEST_NAME = 'test.xlsx'
+TRAIN_NAME = 'train.xlsx'
+
+def get_data(path,name):
+    df = pd.DataFrame()
+    c = 0
+    for file in glob.glob(path):
+        img = io.imread(file)
+        df.insert(c,str(c),img.flatten())
+        c += 1
+
+    df.to_excel(name, index=False)
+
+get_data(TEST_PATH,TEST_NAME)
+get_data(TRAIN_PATH,TRAIN_NAME)

utils/gray_scale.py

@@ -0,0 +1,15 @@
+#convert images to gray and scale to 48x48 pixels
+from skimage import io, color, data
+import glob, cv2
+
+def new_name(name):
+    return name.replace('.png','g.png')
+
+for file in glob.glob('../data/*.png'):
+    img = cv2.imread(file)
+    dsize = (48,48) #new size
+
+    gray_img = color.rgb2gray(img) #conver to grayscale
+    gray_img = cv2.resize(gray_img,dsize, interpolation = cv2.INTER_AREA)
+    gray_img *= 255
+    cv2.imwrite(new_name(file),gray_img)#save image

utils/thresholding.py

@@ -0,0 +1,29 @@
+'''
+Objetive: remove the background and focus the object
+
+Threshold types
+0 - Binary
+1 - Binary Inverted
+2 - Truncated
+3 - Threshold to zero
+4 - Threshold to zero inverted
+'''
+import cv2, glob
+
+BINARY = 0
+THRESHOLD_VALUE = 141
+TEST_PATH = '../data/test/*.png'
+TRAIN_PATH = '../data/train/*.png'
+
+def new_name(name):
+    return name.replace('g.png','.png')
+
+def threshold_image(path):
+    for file in glob.glob(path):
+        image = cv2.imread(file)
+        #apply threshold
+        _, img1 = cv2.threshold(image,THRESHOLD_VALUE, 255, BINARY)
+        cv2.imwrite(new_name(file), img1) #save image
+
+threshold_image(TEST_PATH)
+threshold_image(TRAIN_PATH)