threshold_methods.py

import cv2
import numpy as np
import gradio as gr


def pirahansiah_threshold_method_find_threshold_values_2(grayImg):
    #http://www.jatit.org/volumes/Vol95No21/1Vol95No21.pdf
    #https://pdfs.semanticscholar.org/05b2/d39fce4e8a99897e95f8c75416f65a5a0acc.pdf
    assert grayImg is not None, "file could not be read, check with os.path.exists()"
    #img = cv2.GaussianBlur(self.grayImg, (3, 3), 0)       
    img = grayImg
    # Initialize an array to store the PSNR values for each threshold value
    psnr_values = np.zeros(256)        
    psnr_max=0
    th=0
    # Iterate over all possible threshold values with a step size of 5
    for t in range(0, 256, 5):
        # Threshold the image using the current threshold value
        _, binary = cv2.threshold(img, t, 255, cv2.THRESH_BINARY)            
        # Calculate the PSNR between the binary image and the original image
        psnr = cv2.PSNR(binary, img)            
        # Store the PSNR value
        psnr_values[t] = psnr
        if (psnr_max<psnr):
            psnr_max=psnr
            th=t        
    # Calculate the mean PSNR value
    mean_psnr = np.mean(psnr_values)
    th=int(th/mean_psnr)
    # Find the threshold values that satisfy the condition
    thresh = th #np.argwhere((mean_psnr / k1 < psnr_values) & (psnr_values < mean_psnr / k2)).flatten()
    
    return thresh
def pirahansiah_threshold_method_find_threshold_values_1(grayImg):      
    #https://www.jatit.org/volumes/Vol57No2/4Vol57No2.pdf                      
    assert grayImg is not None, "file could not be read, check with os.path.exists()"
    gray = cv2.GaussianBlur(grayImg, (3, 3), 0)                
    max1=0
    max2=0     
    # Iterate over all possible threshold values
    for t in range(0, 256, 10):            
        # Threshold the image using the current threshold value            
        _, binary = cv2.threshold(gray, t, 255, cv2.THRESH_BINARY)          
        # Find the contours in the binary image            
        contours, hierarchy = cv2.findContours(binary, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)                             
        if max1<=len(contours):
            max1=len(contours)
            max2=t                        
    threshold_values =max2 
    return threshold_values



path  = [['image_0.png']]
inputs_thresh = [
gr.inputs.Image(type="filepath", label="Input Image"),
gr.components.Slider(label="Manual Threshold Value", value=125, minimum=10, maximum=255, step=5),    
gr.inputs.Radio(label="Threshold Methods", 
                choices=[                    
                        "cv2.threshold(grayImg, 128, 255, cv2.THRESH_BINARY)"
                        ,"cv2.threshold(grayImg, 128, 255, cv2.THRESH_BINARY_INV)"
                        ,"cv2.threshold(grayImg, 128, 255, cv2.THRESH_TRUNC)"
                        ,"cv2.threshold(grayImg, 128, 255, cv2.THRESH_TOZERO)"
                        ,"cv2.threshold(grayImg, 128, 255, cv2.THRESH_TOZERO_INV)"
                        ,"cv2.adaptiveThreshold(grayImg, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 11, 2)"
                        ,"cv2.threshold(grayImg, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU,)"
                        ,"Adapted from PirahanSiah Threshold Method I derivative demo"
                        ,"Inspired by PirahanSiah Threshold Method II derivative demo"
                    ]),  
]

outputs_thresh = [
gr.outputs.Image(type="numpy", label="Output Image") 
]

def process_image(input_image, slider_val, radio_choice):
    img = cv2.imread(input_image,0)
    _, binaryImg = cv2.threshold(img, slider_val, 255, cv2.THRESH_BINARY)  
    
    if radio_choice == "cv2.threshold(grayImg, 128, 255, cv2.THRESH_BINARY)":
            _, binaryImg=cv2.threshold(img, 128, 255, cv2.THRESH_BINARY)
    elif radio_choice == "cv2.threshold(grayImg, 128, 255, cv2.THRESH_BINARY_INV)":
            _, binaryImg=cv2.threshold(img, 128, 255, cv2.THRESH_BINARY_INV)
    elif radio_choice == "cv2.threshold(grayImg, 128, 255, cv2.THRESH_TRUNC)":
            _, binaryImg=cv2.threshold(img, 128, 255, cv2.THRESH_TRUNC)
    elif radio_choice == "cv2.threshold(grayImg, 128, 255, cv2.THRESH_TOZERO)":
            _, binaryImg=cv2.threshold(img, 128, 255, cv2.THRESH_TOZERO)
    elif radio_choice == "cv2.threshold(grayImg, 128, 255, cv2.THRESH_TOZERO_INV)":
            _, binaryImg=cv2.threshold(img, 128, 255, cv2.THRESH_TOZERO_INV)
    elif radio_choice == "cv2.adaptiveThreshold(grayImg, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 11, 2)":
            binaryImg=cv2.adaptiveThreshold(img, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 11, 2)
    elif radio_choice == "cv2.threshold(grayImg, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU,)":
           _, binaryImg=cv2.threshold(img, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU,)
    elif radio_choice == "Adapted from PirahanSiah Threshold Method I derivative demo":
           threshval=pirahansiah_threshold_method_find_threshold_values_1(img)        
           _, binaryImg = cv2.threshold(img, threshval, 255, cv2.THRESH_BINARY)                         
    elif radio_choice == "Inspired by PirahanSiah Threshold Method II derivative demo":
           threshval=pirahansiah_threshold_method_find_threshold_values_2(img)        
           _, binaryImg = cv2.threshold(img, threshval, 255, cv2.THRESH_BINARY)                         
    return binaryImg

def on_change(slider_val, radio_choice):
    # Update output
    outputs_thresh[0].update(process_image(
        inputs_thresh[0].value, 
        slider_val, 
        radio_choice)
    )


threshold_methods = gr.Interface(
    fn=process_image,
    inputs=inputs_thresh,
    outputs=outputs_thresh, 
    on_change=on_change,
    examples=path,
    title="Computer Vision and Deep Learning by Farshid PirahanSiah",
    live=True
    )

  



# class Thresholding:
#     def __init__(self, grayImg):
#         self.grayImg = grayImg

#     def manually_python(self):        
#         threshval = 128
#         binaryImg = np.where(self.grayImg < threshval, self.grayImg, 0) if threshval < 128 else np.where(self.grayImg > threshval, self.grayImg, 0)
#         return binaryImg
    
#     def manually(self,threshval):        
#         binaryImg = np.zeros_like(self.grayImg)
#         for i in range(self.grayImg.shape[0]): #height 
#             for j in range(self.grayImg.shape[1]): #width 
#                 if self.grayImg[i, j] < threshval:
#                     binaryImg[i, j] = 0 
#                 else:
#                     binaryImg[i, j] = 1 
#         return binaryImg    

#     def pirahansiah_threshold_method_find_threshold_values_2(self):
#         #http://www.jatit.org/volumes/Vol95No21/1Vol95No21.pdf
#         #https://pdfs.semanticscholar.org/05b2/d39fce4e8a99897e95f8c75416f65a5a0acc.pdf
#         assert self.grayImg is not None, "file could not be read, check with os.path.exists()"
#         #img = cv2.GaussianBlur(self.grayImg, (3, 3), 0)       
#         img = self.grayImg
#         # Initialize an array to store the PSNR values for each threshold value
#         psnr_values = np.zeros(256)        
#         psnr_max=0
#         th=0
#         # Iterate over all possible threshold values with a step size of 5
#         for t in range(0, 256, 5):
#             # Threshold the image using the current threshold value
#             _, binary = cv2.threshold(img, t, 255, cv2.THRESH_BINARY)            
#             # Calculate the PSNR between the binary image and the original image
#             psnr = cv2.PSNR(binary, img)            
#             # Store the PSNR value
#             psnr_values[t] = psnr
#             if (psnr_max<psnr):
#                 psnr_max=psnr
#                 th=t        
#         # Calculate the mean PSNR value
#         mean_psnr = np.mean(psnr_values)
#         th=int(th/mean_psnr)
#         # Find the threshold values that satisfy the condition
#         thresh = th #np.argwhere((mean_psnr / k1 < psnr_values) & (psnr_values < mean_psnr / k2)).flatten()
        
#         return thresh
#     def pirahansiah_threshold_method_find_threshold_values_1(self):      
#         #https://www.jatit.org/volumes/Vol57No2/4Vol57No2.pdf                      
#         assert self.grayImg is not None, "file could not be read, check with os.path.exists()"
#         gray = cv2.GaussianBlur(self.grayImg, (3, 3), 0)                
#         max1=0
#         max2=0     
#         # Iterate over all possible threshold values
#         for t in range(0, 256, 10):            
#             # Threshold the image using the current threshold value            
#             _, binary = cv2.threshold(gray, t, 255, cv2.THRESH_BINARY)          
#             # Find the contours in the binary image            
#             contours, hierarchy = cv2.findContours(binary, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)                             
#             if max1<=len(contours):
#                 max1=len(contours)
#                 max2=t                        
#         threshold_values =max2 
#         return threshold_values


#     def opencv_th(self):
#         font = cv2.FONT_HERSHEY_SIMPLEX
#         fontScale = 2
#         color = (0, 0, 0)
#         colorInv = (255, 255, 255)
#         thickness = 2
#         # Set the position of the text
#         textX = 25
#         textY = 45
#         textSize, _ = cv2.getTextSize("Otsu Method   ", font, fontScale, thickness)
#         # Draw a white rectangle behind the text
        
#         # Apply different thresholding methods
#         _, binaryImg = cv2.threshold(self.grayImg, 128, 255, cv2.THRESH_BINARY)                
#         cv2.rectangle(binaryImg, (textX, textY - textSize[1]), (textX + textSize[0], textY), (255, 255, 255), -1)
#         cv2.putText(binaryImg, 'Binary', (textX, textY), font, fontScale, color, thickness)
        
#         _, binaryInvImg = cv2.threshold(self.grayImg, 128, 255, cv2.THRESH_BINARY_INV)
#         cv2.rectangle(binaryInvImg, (textX, textY - textSize[1]), (textX + textSize[0], textY), (255, 255, 255), -1)
#         cv2.putText(binaryInvImg, 'Binary Inv', (textX, textY), font, fontScale, color, thickness)
        
#         _, truncImg = cv2.threshold(self.grayImg, 128, 255, cv2.THRESH_TRUNC)
#         cv2.rectangle(truncImg, (textX, textY - textSize[1]), (textX + textSize[0], textY), (255, 255, 255), -1)
#         cv2.putText(truncImg, 'Trunc', (textX, textY), font, fontScale, color, thickness)
        
#         _, toZeroImg = cv2.threshold(self.grayImg, 128, 255, cv2.THRESH_TOZERO)
#         cv2.rectangle(toZeroImg, (textX, textY - textSize[1]), (textX + textSize[0], textY), (255, 255, 255), -1)
#         cv2.putText(toZeroImg, 'To Zero', (textX, textY), font, fontScale, color, thickness)
        
#         _, toZeroInvImg = cv2.threshold(self.grayImg, 128, 255, cv2.THRESH_TOZERO_INV)
#         cv2.rectangle(toZeroInvImg, (textX, textY - textSize[1]), (textX + textSize[0], textY), (255, 255, 255), -1)
#         cv2.putText(toZeroInvImg, 'To Zero Inv', (textX, textY), font, fontScale, color, thickness)
        
#         adaptiveImg = cv2.adaptiveThreshold(self.grayImg, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 11, 2)
#         cv2.rectangle(adaptiveImg, (textX, textY - textSize[1]), (textX + textSize[0], textY), (255, 255, 255), -1)
#         cv2.putText(adaptiveImg, 'Adaptive', (textX, textY), font, fontScale, color, thickness)
        
#         otsu_threshold, image_result = cv2.threshold(self.grayImg, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU,)        
#         cv2.rectangle(image_result, (textX, textY - textSize[1]), (textX + textSize[0], textY), (255, 255, 255), -1)
#         cv2.putText(image_result, 'Otsu Threshold Method', (textX, textY), font, fontScale, color, thickness)

#         threshval=self.pirahansiah_threshold_method_find_threshold_values_1()        
#         th_img = th.manually(threshval)
#         binaryImg = th_img * 255
#         cv2.rectangle(binaryImg, (textX, textY - textSize[1]), (textX + textSize[0], textY), (255, 255, 255), -1)
#         cv2.putText(binaryImg, 'PirahanSiah Threshold ', (textX, textY), font, fontScale, color, thickness)


#         cv2.rectangle(self.grayImg, (textX, textY - textSize[1]), (textX + textSize[0], textY), (255, 255, 255), -1)
#         cv2.putText(self.grayImg, 'Original', (textX, textY), font, fontScale, color, thickness)
#         # Concatenate the images into a grid with 3 rows and 3 columns
#         row1 = np.concatenate((self.grayImg, binaryImg, binaryInvImg), axis=1)
#         row2 = np.concatenate((truncImg, toZeroImg, toZeroInvImg), axis=1)                
#         row3 = np.concatenate((adaptiveImg, image_result, binaryImg), axis=1) # np.zeros_like(adaptiveImg)
#         concatenatedImg = np.concatenate((row1, row2, row3), axis=0)            
#         # Resize the concatenated image to fit the screen resolution
#         screenRes = (1920-200, 1080-200)
#         scaleWidth = screenRes[0] / concatenatedImg.shape[1]
#         scaleHeight = screenRes[1] / concatenatedImg.shape[0]
#         scale = min(scaleWidth, scaleHeight)
#         windowWidth = int(concatenatedImg.shape[1] * scale)
#         windowHeight = int(concatenatedImg.shape[0] * scale)
#         resizedImg = cv2.resize(concatenatedImg, (windowWidth, windowHeight))
#         # Display the resized image
#         cv2.imshow('Thresholded Images', resizedImg)
#         cv2.waitKey(0)
#         cv2.destroyAllWindows()


# if __name__ == "__main__":    
#     img = cv2.imread("opencv.png")    
#     gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)    
#     th = Thresholding(gray)    
#     th.opencv_th()

    #threshval = 128
    # read an image
    # convert it to grayscale
    #threshval=th.pirahansiah_threshold_method_find_threshold_values_1()
    # threshval=th.pirahansiah_threshold_method_find_threshold_values_2()        
    # th_img = th.manually(threshval)
    # binaryImg = th_img * 255
    # cv2.imshow('image', binaryImg)
    # cv2.waitKey(100)


'''
cv::Mat binaryImg = threshval < 128 ? (grayImg < threshval) : (grayImg > threshval);
#thresholding #opencv #python

The PirahanSiah’s method for thresholding, described in the paper, uses a gray-scale histogram, 
thresholding range, and the Peak Signal-to-Noise Ratio (PSNR) to segment images and find the best 
threshold values to binarize the image. They argue that thresholding is an important problem in 
pattern recognition and use the PSNR quality measure to assess the similarities between the 
original and binarized image. They calculate PSNRs for every threshold value and use the 
difference between the PSNR of the previous threshold image and the new one to select the 
threshold value. They also describe a multi-threshold algorithm that applies multiple 
threshold values and computes the total number of blobs or objects in an image for each threshold.
The peak threshold values are those with the highest total number of blobs compared to their threshold neighbors.
In addition, their method uses thresholding on images suitable for OCR systems, LPR systems, etc.

The proposed adaptive threshold method, based on the Peak Signal-to-Noise Ratio (PSNR), 
has the potential to be applied in all domains, such as LPR and OCR. The proposed algorithm 
achieves competitive results in four databases, including Malaysian vehicle, standard, 
printed and handwritten images. The objective of this research was to develop a new single 
adaptive thresholding algorithm that works for a wide range of pattern recognition applications. 
The proposed method has been implemented in four different types of applications and compared 
with other methods. The results show that the proposed algorithm achieves the objective because 
it has obtained reasonable results in all four areas/domains.
https://www.jatit.org/volumes/Vol57No2/4Vol57No2.pdf 
http://www.jatit.org/volumes/Vol95No21/1Vol95No21.pdf
https://pdfs.semanticscholar.org/05b2/d39fce4e8a99897e95f8c75416f65a5a0acc.pdf
'''