First, introduce the imagehash.

To date, when encountering duplicates, we would remove them or put them in a same fold to denoise.

• 

• 

Here is the result, the similarity matrix

array([[0.       , 1.       , 0.6328125],
       [1.       , 0.       , 0.6328125],
       [0.6328125, 0.6328125, 0.       ]])

---

display duplicate images

# show 20 duplicate images


count = 20
tmp = 0

pairs = {}
for row, col in zip(*duplicates):
    # diagonal
    if row == col:
        continue

    path1 = paths[row]
    path2 = paths[col]
    print(path1)
    print(path2)
    print(sims2[row, col])

    image1 = cv2.imread(path1)
    image2 = cv2.imread(path2)

    if image1.shape[0] > image1.shape[1] / 2:
        fig,ax = plt.subplots(figsize=(20,20), ncols=2)
    elif image1.shape[1] > image1.shape[0] / 2:
        fig,ax = plt.subplots(figsize=(20,20), nrows=2)
    else:
        fig,ax = plt.subplots(figsize=(20,30), nrows=2)
    ax[0].imshow(image1)
    ax[1].imshow(image2)
    plt.show()
    
    tmp += 1
    if tmp > count:
        break