Update app.py

Image Preprocessing: Enhance contrast and apply grayscale to improve text extraction.

Custom Regex: Adjust regex to capture the serial number format seen in the image.

Alternative Models: Consider using easyocr for better performance on printed meter readings.

app.py

@@ -1,46 +1,58 @@
 import gradio as gr
 from transformers import TrOCRProcessor, VisionEncoderDecoderModel
-from PIL import Image
+from PIL import Image, ImageEnhance, ImageOps
 import torch
 import re
+import easyocr
+
+# Initialize EasyOCR Reader
+reader = easyocr.Reader(['en'])
 
 # Load the pre-trained model and processor
 processor = TrOCRProcessor.from_pretrained('microsoft/trocr-small-printed')
 model = VisionEncoderDecoderModel.from_pretrained('microsoft/trocr-small-printed')
 model.eval()
 
+def enhance_image(image):
+    # Enhance the image for better OCR results
+    image = ImageEnhance.Contrast(image).enhance(2.0)
+    image = ImageOps.grayscale(image)
+    return image
+
+
 def extract_details(text):
     # Use regex to extract serial number and meter reading
-    sr_no = re.search(r'(?:Sr\.No|SR\.NO|Serial Number|SR NO)[:\s]*([\w\d\-]+)', text, re.IGNORECASE)
-    meter_reading = re.search(r'(?:Reading|Meter Reading|Reading:)[:\s]*([\d\.]+)', text, re.IGNORECASE)
+    sr_no = re.search(r'(?:Sr\.No|SR\.NO|Serial Number|SR NO)[:\s]*([\d]+)', text, re.IGNORECASE)
+    meter_reading = re.search(r'(?:Reading|Meter Reading|kWh|KW H)[:\s]*([\d]+)', text, re.IGNORECASE)
     sr_no_text = sr_no.group(1) if sr_no else "Not Found"
     reading_text = meter_reading.group(1) if meter_reading else "Not Found"
     return sr_no_text, reading_text
 
+
 def extract_meter_reading(image):
     try:
-        # Convert and preprocess the image
-        image = Image.fromarray(image).convert("RGB")
-        pixel_values = processor(images=image, return_tensors="pt").pixel_values
+        # Enhance image
+        image = Image.fromarray(image)
+        enhanced_image = enhance_image(image)
 
-        # Generate text from image
-        with torch.no_grad():
-            generated_ids = model.generate(pixel_values)
-        generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
+        # Use EasyOCR to get better results
+        result = reader.readtext(enhanced_image)
+        text = " ".join([item[1] for item in result])
 
-        # Extract details from the generated text
-        sr_no, reading = extract_details(generated_text)
+        # Extract details from the OCR text
+        sr_no, reading = extract_details(text)
         return f"Serial Number: {sr_no}\nMeter Reading: {reading}"
     except Exception as e:
         return f"Error: {str(e)}"
 
+
 # Create Gradio interface
 iface = gr.Interface(
     fn=extract_meter_reading,
     inputs=gr.Image(type="numpy", label="Upload or Capture Meter Image"),
     outputs="text",
     title="Meter Reading and Serial Number Extractor",
-    description="Upload a meter image to extract the serial number and meter reading using a lightweight OCR model."
+    description="Upload a meter image to extract the serial number and meter reading using OCR techniques."
 )
 
 # Launch the Gradio app