Upload app.py

app.py

@@ -0,0 +1,141 @@
+"""
+Jialu Bi
+CS5330 Lab 1
+Generator for ASCII art
+"""
+
+import cv2
+import numpy as np
+import matplotlib.pyplot as plt
+import gradio as gr
+from skimage.metrics import structural_similarity as ssim
+from skimage.metrics import mean_squared_error
+
+# A class for ASCII genrator related functions
+class ASCIIGenerator:
+    def __init__(self):
+        self.ASCIIChars = ['#', '@', '%', '*', '+', '-', '.', ' ']
+        # self.ASCIIChars = ['█', '▓', '▒', '░', ' ']
+
+    # Resize the image to make sure consistent result
+    def resizeImg(self, img):
+        h, w, c = img.shape
+        aspectRatio = w / h
+        imgResized = cv2.resize(img, (int(160 * aspectRatio), 160))
+        return imgResized
+
+    # Convert to gray scale
+    # The input should be the resized image
+    def cvtGrayScale(self, imgResized):
+        R, G, B = cv2.split(imgResized)
+        R, G, B = R.astype(np.float32), G.astype(np.float32), B.astype(np.float32)
+
+        # Apply weights for more refined control
+        imgGray = 0.299 * R + 0.587 * G + 0.114 * B
+        imgGray = np.clip(imgGray, 0, 255).astype(np.uint8)
+        return imgGray
+
+    # Preprocess the image
+    # The input should be the gray scale image
+    def preprocess(self, imgGray):
+        # Histogram equalization
+        clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
+        imgEqualized = clahe.apply(imgGray)
+
+        # Increase contrast
+        imgEqualized = cv2.convertScaleAbs(imgEqualized, alpha=1.6, beta=0)
+
+        # Detect the edges and emphasize them
+        edges = cv2.Canny(imgEqualized, threshold1=50, threshold2=150)
+        invertedEdges = cv2.bitwise_not(edges)
+        imgEnhanced = cv2.addWeighted(imgGray, 0.7, invertedEdges, 0.3, 0)
+        return imgEnhanced
+
+    # Map to ASCII Characters
+    # The input should be the preprocessed image
+    def mapASCII(self, imgEnhanced):
+        ASCIIArt = ''
+        h, w = imgEnhanced.shape
+
+        # Iterate through the pixels to create the ASCII art
+        for i in range(h):
+            for j in range(w):
+                pixelVal = imgEnhanced[i, j]
+                # Assign a proper ASCII character according to the pixel's intensity value
+                ASCIIArt += self.ASCIIChars[pixelVal * len(self.ASCIIChars) // 256]
+            ASCIIArt += '\n'
+        return ASCIIArt
+
+    # Main function for generating ASCII art
+    def generateASCIIArt(self, img):
+        imgResized = self.resizeImg(img)
+        imgGray = self.cvtGrayScale(imgResized)
+        imgEnhanced = self.preprocess(imgGray)
+        ASCIIArt = self.mapASCII(imgEnhanced)
+        return ASCIIArt
+
+    # Convert ASCII art back to grayscale for performance evaluation
+    def ASCIItoGray(self, ASCIIArt):
+        # Create a dictionary to store the corresponding intensity of ASCII characters
+        ASCIICharsVal = {}
+        offset = 0
+        for char in self.ASCIIChars:
+            ASCIICharsVal[char] = offset
+            offset += 256 // len(self.ASCIIChars)
+
+        # Map ASCII characters back to pixels
+        ASCIIGray = []
+        lines = ASCIIArt.split('\n')[:-1]
+        for line in lines:
+            row = []
+            for char in line:
+                row.append(ASCIICharsVal[char])
+            ASCIIGray.append(row)
+        ASCIIGray = np.array(ASCIIGray, dtype=np.uint8)
+
+        return ASCIIGray
+
+    # Evaluate the similarity between the grayscale version of the original image and its ASCII art version
+    def compare(self, ASCIIGray, imgGray):
+        # Compute SSIM
+        ssimVal, _ = ssim(imgGray, ASCIIGray, full=True)
+        # Compute MSE
+        mseVal = mean_squared_error(imgGray, ASCIIGray)
+        # Compute PSNR
+        psnrVal = cv2.PSNR(imgGray, ASCIIGray)
+        return ssimVal, mseVal, psnrVal
+
+
+# Gradio interface function
+def generateArt(img):
+    if img is None:
+        return None, None, None
+
+    # Generate the ASCII art
+    generator = ASCIIGenerator()
+    ASCIIArt = generator.generateASCIIArt(img)
+
+    # Evaluate the performance of the algorithm
+    imgResized = generator.resizeImg(img)
+    imgGray = generator.cvtGrayScale(imgResized)
+    ASCIIGray = generator.ASCIItoGray(ASCIIArt)
+    scores = generator.compare(ASCIIGray, imgGray)
+    report = f'SSIM: {round(scores[0], 2)}, MSE: {round(scores[1], 2)}, PSNR: {round(scores[2], 2)}'
+
+    # Save the ASCII art to a text file
+    outputPath = 'ascii_art.txt'
+    with open(outputPath, 'w', encoding='utf-8') as file:
+        file.write(ASCIIArt)
+    return f'<pre style="font-size: 8px; line-height: 0.62;">{ASCIIArt}</pre>', outputPath, report
+
+# Create a Gradio interface
+with gr.Blocks() as demo:
+    gr.Markdown('## ASCII Art Generator\nUpload an image to generate ASCII art.')
+    imgInput = gr.Image(type='numpy', label='Upload Image')
+    ASCIIOutput = gr.HTML(label='ASCII Art Output')
+    fileOutput = gr.File(label='Download ASCII Art')
+    scoreOutput = gr.Textbox(label='ASCII Conversion Performance')
+    imgInput.change(fn=generateArt, inputs=imgInput, outputs=[ASCIIOutput, fileOutput, scoreOutput])
+
+
+demo.launch(share=True)