Upload

Author: khalidmoataz

Data_preprocessing/pre_process_data.py

@@ -0,0 +1,21 @@
+VGG_Model = load_model('vgg_4096.h5',compile=False)
+
+def pre_process_data(the_path,model):
+    images_features = []
+    img_array = []
+    Rgb_Colors = []
+    for path, _, files in os.walk(the_path):
+        for file in files:
+            img_path = path + str('\\') + file
+            img = cv2.imread(img_path)
+            img = cv2.resize(img,(224,224))
+            gimg = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+            img_data = img
+            img_array.append(gimg)
+            Rgb_Colors.append(img_data)
+            img_data = np.expand_dims(img_data, axis=0)
+            img_data = preprocess_input(img_data)
+
+            images_features.append(model.predict(img_data))
+
+    return img_array,images_features,Rgb_Colors 

Data_preprocessing/process_query_image.py

@@ -0,0 +1,33 @@
+def process_query_image(img_path,images_features,img_array,Rgb_Colors,VGG_Model): # lut
+    windowsize_r = 28
+    windowsize_c = 28
+    yy = np.array(images_features)
+    yy = yy.reshape((len(images_features),4096))
+    
+    img = cv2.imread(img_path)
+    img = cv2.resize(img, (224,224))
+    gimg = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+
+    img_data = img
+    new = []
+    test_image = gimg
+    test_image = test_image.reshape(224,224,)
+    # Histogram of Query image
+    for r in range(0,test_image.shape[0] - windowsize_r, windowsize_r):
+        for c in range(0,test_image.shape[1] - windowsize_c, windowsize_c):
+            window = test_image[r:r+windowsize_r,c:c+windowsize_c]
+            new.append(cv2.calcHist(window,[0,1,2],None,[8,8,8],[0, 256, 0, 256, 0, 256]))    
+
+    
+    img_data = np.expand_dims(img_data, axis=0)
+    gimg = np.expand_dims(gimg, axis=0)
+    
+    Rgb = np.append(Rgb_Colors,img_data, axis=0)
+    the_images = np.append(img_array, gimg, axis=0)
+
+    img_data = preprocess_input(img_data)
+
+    # Extract Features Query
+    new_im_pred = VGG_Model.predict(img_data)
+    preds1 = np.append(yy, new_im_pred, axis=0)
+    return preds1,new,the_images,Rgb

Get_Similar_Images/Display Results.py

@@ -0,0 +1,31 @@
+def display_results_after_LCH(the_images,results,nearest_neighbors):
+    imgg = the_images[results[0][1]]
+    plt.title('Query Image')
+    plt.imshow(cv2.cvtColor(imgg, cv2.COLOR_BGR2RGB))
+    fig = plt.figure("Re")
+    fig.set_figheight(25)
+    fig.set_figwidth(25)
+    k=1
+    for i in range(1,5):
+        imgg = the_images[results[i][1]]
+        imgg = imgg.reshape(224,224,3)
+        ax = fig.add_subplot(1, 6, k)
+        plt.imshow(cv2.cvtColor(imgg, cv2.COLOR_BGR2RGB))
+        plt.axis("off")
+        k=k+1
+
+def display_results_before_LCH(the_images,results,nearest_neighbors):
+    imgg = the_images[nearest_neighbors[0]]
+    plt.title('Query Image')
+    plt.imshow(cv2.cvtColor(imgg, cv2.COLOR_BGR2RGB))
+    zfig = plt.figure("Re")
+    zfig.set_figheight(25)
+    zfig.set_figwidth(25)
+    k=1
+    for i in range(1,5):
+        imgg = the_images[nearest_neighbors[i]]
+        imgg = imgg.reshape(224,224,3)
+        ax = zfig.add_subplot(1, 6, k)
+        plt.imshow(cv2.cvtColor(imgg, cv2.COLOR_BGR2RGB))
+        plt.axis("off")
+        k=k+1

Get_Similar_Images/get_nearest_neighbor.py

@@ -0,0 +1,26 @@
+def get_nearest_neighbor_and_similarity(preds1, K):
+    dims = 4096
+    n_nearest_neighbors = K+1
+    # Specify no of trees
+    trees = 1000
+    file_index_to_file_vector = {}
+
+    # build ann index
+    t = AnnoyIndex(dims)
+    for i in range(preds1.shape[0]):
+
+        file_vector = preds1[i]
+        file_index_to_file_vector[i] = file_vector
+        t.add_item(i, file_vector)
+    # Build trees
+    t.build(trees)
+
+    # Get NN
+    for i in range(preds1.shape[0]):
+        master_vector = file_index_to_file_vector[i]
+
+        named_nearest_neighbors = []
+        nearest_neighbors = t.get_nns_by_item(i, n_nearest_neighbors)
+
+    return nearest_neighbors
+

Get_Similar_Images/get_similar_images.py

@@ -0,0 +1,32 @@
+def get_similar_images(nearest_neighbors,images1,query_hist):
+    histr = []
+    temp = []
+    windowsize_r = 28
+    windowsize_c = 28
+    for i in nearest_neighbors:
+        test_image = images1[i]        
+        test_image = test_image.reshape(224,224,)
+        for r in range(0,test_image.shape[0] - windowsize_r, windowsize_r):
+            for c in range(0,test_image.shape[1] - windowsize_c, windowsize_c):
+                window = test_image[r:r+windowsize_r,c:c+windowsize_c]
+                temp.append(cv2.calcHist(window,[0,1,2],None,[8,8,8],[0, 256, 0, 256, 0, 256]))
+        histr.append(temp)
+        temp = []
+    the_sum = 0
+    results = {}
+    print(len(histr[0]))
+    for y in range(0,5):
+        the_sum = 0
+        for u in range(0,49):
+            f = np.array(query_hist[u])
+            h = np.array(histr[y][u])
+
+            f = f.flatten()
+            h = h.flatten()
+
+            s = dist.euclidean(f,h)
+            the_sum =  the_sum + s
+        results[nearest_neighbors[y]] = the_sum
+    results = sorted([(v, k) for (k, v) in results.items()], reverse = False)
+    print(results)
+    return results

Runner.py

@@ -0,0 +1,24 @@
+from keras.preprocessing import image
+from keras.applications.vgg16 import VGG16
+from keras.applications.vgg16 import preprocess_input
+import numpy as np
+import cv2
+import os
+import annoy
+from annoy import AnnoyIndex
+from keras.models import load_model
+from scipy.spatial import distance as dist
+import matplotlib.pyplot as plt
+
+
+from Data_preprocessing import *
+
+from Get_Similar_Images import *
+
+
+img_array,images_features,Rgb_colors = pre_process_data('C:\\Users\\khali\\101_ObjectCategories\\random',VGG_Model)
+preds1,new,the_images,Rgb = process_query_image('C:\\Users\\khali\\cam.jpg',images_features,img_array,Rgb_colors,VGG_Model)
+nearest_neighbors = get_nearest_neighbor_and_similarity(preds1,15)
+results = get_similar_images(nearest_neighbors,the_images,new)
+display_results_before_LCH(Rgb,results,nearest_neighbors)
+display_results_after_LCH(Rgb,results,nearest_neighbors)