Application File

app.py

@@ -0,0 +1,90 @@
+import streamlit as st
+import numpy as np
+import matplotlib.pyplot as plt
+import scipy.fftpack as fp
+import cv2
+from PIL import Image
+
+
+# ideal filter
+def ideal_filter(rows, cols, D0, filtr):
+    H = np.zeros(shape = (rows, cols))
+    for i in range(rows):
+        for j in range(cols):
+            # euclidean distance from u,v to origin of frequency
+
+            Duv = np.sqrt(np.power(i - rows/2, 2) + np.power(j - cols/2, 2)) 
+            if Duv < D0:
+                H[i,j] = 1.0
+    if filtr == "High Pass":
+        H = 1-H
+    #H = H*255
+    cv2.imwrite('filter.jpg',np.abs(H*255))
+    return H
+
+def butterworth_filter(rows, cols, n_order, D0, filtr):
+    H = np.zeros(shape = (rows, cols))
+    for i in range(rows):
+        for j in range(cols):
+        
+            Duv = np.sqrt(np.power(i - rows/2, 2) + np.power(j - cols/2, 2))
+            H[i,j] = 1/(1+((Duv/D0)**(2*n_order)))
+    #import pdb;pdb.set_trace()
+    if filtr == "High Pass":
+        H = 1-H
+    
+    cv2.imwrite('filter.jpg',np.abs(H*255))
+    
+    return H
+
+def gaussian_filter(rows, cols, filtr):
+    #import pdb;pdb.set_trace()
+    H = np.zeros(shape = (rows, cols))
+    for i in range(rows):
+        for j in range(cols):
+            Duv = np.sqrt(np.power(i - rows/2, 2) + np.power(j - cols/2, 2))
+            H[i,j] = np.exp(-((Duv**2)/(2*(D0**2))))
+    if filtr == "High Pass":
+        H = 1-H
+    #H = H*255
+    cv2.imwrite('filter.jpg',np.abs(H*255))
+    return H
+
+
+
+
+uploaded_file = st.sidebar.file_uploader("Upload image", type = ["jpeg", "jpg", "png"])
+filtr = st.sidebar.radio("Filters", ("Low Pass", "High Pass"))
+kernel = st.sidebar.radio("Kernels", ("Ideal", "Butterworth", "Gaussian"))
+D0 = st.sidebar.slider("Cutoff Frequency", min_value = 0, max_value = 75)
+n_order = st.sidebar.number_input(label = "Order", min_value = 0, max_value = 5)
+
+if uploaded_file is not None:
+    img = Image.open(uploaded_file)
+    img.save("read_image.jpg")
+    st.write("Source Image")
+    st.image("read_image.jpg", width = 300)
+    img = cv2.imread("read_image.jpg", 0)
+    rows, cols = img.shape
+    if kernel == "Ideal":
+        H = ideal_filter(rows, cols, D0, filtr)
+    elif kernel == "Gaussian":
+        H = gaussian_filter(rows, cols, filtr)
+    elif kernel == "Butterworth":
+        H = butterworth_filter(rows, cols, n_order, D0, filtr)
+    
+    H = fp.fft2(fp.ifftshift(H)) # fast fourier transform
+    f_img = fp.fft2(img) # fast fourier transform
+    conv_img = np.multiply(H, f_img)
+
+    inv_img = fp.ifft2(conv_img).real
+
+    output_img = ((inv_img - np.min(inv_img))/np.max(inv_img))*255
+
+    #plt.imshow(output_img, cmap='gray')
+    st.write("Kernel")
+    st.image("filter.jpg", width = 300)
+    cv2.imwrite('output_image.jpg',output_img)
+    st.write(f"Target Image with {filtr} {kernel} Filter")
+    st.image("output_image.jpg", width = 300)
+