Hi, I dont find the function "Contrast Limited Adaptive Histogram Equalization (CLAHE)"! Where is it? Please

Hi, Thank you for reaching out. We did not include the CLAHE method to the main pipeline. Rather we applied CLAHE on the whole dataset and made a separate instance so that we don't have to perform CLAHE everytime the image is loaded. However, I added the function to the repo in this link: enhancements.py

Please look for the function applyCLAHE. `

def applyCLAHE(image, display: bool = False):

    image_bw = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

    # The declaration of CLAHE
    # clipLimit -> Threshold for contrast limiting
    clahe = cv2.createCLAHE(clipLimit=5)
    final_img = clahe.apply(image_bw)  # + 30
    final_img = np.stack((final_img,) * 3, axis=-1)

    # Ordinary thresholding the same image
    _, ordinary_img = cv2.threshold(image_bw, 155, 255, cv2.THRESH_BINARY)

    if display:
        fig = plt.figure(figsize=(9, 3), dpi=300)
        rows, cols = 1, 3

        # Display the original image
        fig.add_subplot(rows, cols, 1)
        plt.imshow(image, cmap=plt.cm.bone);
        plt.axis('off')
        plt.title("Input Image")

        # Display the thresholded image
        fig.add_subplot(rows, cols, 2)
        plt.imshow(ordinary_img, cmap=plt.cm.gray);
        plt.axis('off')
        plt.title("Binary Thresholded Image")

        # Display the CLAHE processed image
        fig.add_subplot(rows, cols, 3)
        plt.imshow(final_img, cmap=plt.cm.gray)
        plt.axis('off')
        plt.title("CLAHE Image")
        plt.show()
    # print(final_img.shape)
    return final_img

Please look for the function applyCLAHE. `

def applyCLAHE(image, display: bool = False):

    image_bw = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

    # The declaration of CLAHE
    # clipLimit -> Threshold for contrast limiting
    clahe = cv2.createCLAHE(clipLimit=5)
    final_img = clahe.apply(image_bw)  # + 30
    final_img = np.stack((final_img,) * 3, axis=-1)

    # Ordinary thresholding the same image
    _, ordinary_img = cv2.threshold(image_bw, 155, 255, cv2.THRESH_BINARY)

    if display:
        fig = plt.figure(figsize=(9, 3), dpi=300)
        rows, cols = 1, 3

        # Display the original image
        fig.add_subplot(rows, cols, 1)
        plt.imshow(image, cmap=plt.cm.bone);
        plt.axis('off')
        plt.title("Input Image")

        # Display the thresholded image
        fig.add_subplot(rows, cols, 2)
        plt.imshow(ordinary_img, cmap=plt.cm.gray);
        plt.axis('off')
        plt.title("Binary Thresholded Image")

        # Display the CLAHE processed image
        fig.add_subplot(rows, cols, 3)
        plt.imshow(final_img, cmap=plt.cm.gray)
        plt.axis('off')
        plt.title("CLAHE Image")
        plt.show()
    # print(final_img.shape)
    return final_img

Thank you for responding. Let me further ask that:

1/ "PlantVillage-Tomato.zip" that I downloaded from https://data.mendeley.com/public-files/datasets/tywbtsjrjv/files/d5652a28-c1d8-4b76-97f3-72fb80f94efc/file_downloaded" has CLAHE applied?

2/ According to your article "Before applying CLAHE, the leaf image was converted from RGB color space to Hunter Lab color space" while I see the CLAHE function you provided is BGR to Gray color space. Which is correct for you?

3/ The article says "The intensity channel of the leaf image was divided into P ×Q regions" and "a value of 7 for both P and Q provided the best results.” I can't find where you have set P=Q=7 in the applyCLAHE function?

4/ In the preprocessing step, you apply CLAHE on the whole dataset. So I understand the steps you do correctly: 1- applyHistogramEqualization, 2-applyHEFFilter, 3- applyCLAHE, 4. train model (Data Augmentation during runtime).

5/ According to Ablation study of different components of the proposed pipeline (applied CLAHE, Augmentation, & Classifier Network with Accuracy 99.30). Can you provide this model? I want to test it and compare it with the model I created from the training results according to your instructions. (I just found: MobileNetV1_WithoutCLAHE_NoAug_WithoutDense_ValBest.h5, MobileNetV2_WithCLAHE_NoAug_WithoutDense_ValBest.h5 instead of what I expected is MobileNetV2_WithCLAHE_withAug_withDense_ValBest.h5)

I greatly appreciate your article and contribution. Thank you again

redwankarimsony / project-tomato

Hi, I dont find the function "Contrast Limited Adaptive Histogram Equalization (CLAHE)"! Where is it? Please #4