17-03-2025

Image Pre-processing

def pre_process_image(image_path, contrast_factor=1.5):
 
# Load image
 
image = Image.open(image_path).convert("RGB")
 
# Increase contrast
 
enhancer = ImageEnhance.Contrast(image)
 
image = enhancer.enhance(contrast_factor)
 
np_image = np.array(image)
 
# Create a circular mask
 
height, width, _ = np_image.shape
 
center = (370, 305)
 
radius = 100
mask = Image.new("L", (width, height), 0)
 
draw = ImageDraw.Draw(mask)
 
draw.ellipse((center[0] - radius, center[1] - radius, center[0] + radius, center[1] + radius), fill=255)
# Apply the mask to the image
masked_image = Image.fromarray(np_image)
masked_image.putalpha(mask)
 
masked_image = np.array(masked_image)
 
# Crop the image to the diameter of the circle
 
left = center[0] - radius
 
top = center[1] - radius
 
right = center[0] + radius
 
bottom = center[1] + radius
 
masked_image = masked_image[top:bottom, left:right]
 
  
 
# Extract RGB channels and alpha channel
 
np_image_rgb = masked_image[:, :, :3]
 
alpha_channel = masked_image[:, :, 3]
 
  
 
# Create a black background
 
black_background = np.zeros_like(np_image_rgb)
 
  
 
# Combine the RGB image with the black background using the alpha channel
 
np_image_rgb = np.where(alpha_channel[..., None] == 0, black_background, np_image_rgb)
 
  
  
 
return np_image_rgb

Example

Still some limitations using SAM2

Example

Other Pre-processing steps tried that didn’t make much of a difference

• increasing contrast
• increasing brightness
• preliminary edge-detection

Pre-Image Subtraction

sub_image_filename = "../data/trial-1/10-A10_MQ_Ket_1_DS_PREIMAGE.jpg"
 
np_subtraction = pre_process_image(sub_image_filename, contrast_factor=1)
 
plt.imshow(np_subtraction)
 
  
 
# Ensure both images are of the same shape
 
if np_image.shape == np_subtraction.shape:
	
	# Subtract np_image from np_subtraction
	
	result_image = np_image[:, :, :3] - np_subtraction[:, :, :3]
	
	result_image = np.dstack((result_image, np_image[:, :, 3]))
	
	  
	
	# Display the result
	
	plt.figure(figsize=(10, 10))
	
	plt.imshow(result_image)
	
	plt.title("Subtracted Image")
	
	plt.axis('off')
	
	plt.show()
 
else:
 
	print("The images are not of the same shape and cannot be subtracted.")

Example

Looked really promising, it should be relatively easy for a model to learn given the current image and the pre-image

Testing a New Model: CV2_MOG2

Advantagous

• Can set background using learning rates
• Can ensure the background doesn’t update
• Uses videos/image sequences by default

Some Gifs

• Determine if some wells are under-performing
  • Get a quick fit metric, well position
• Try to extract useful information now
• Higher cross-linking
  • tighter packed
  • brighter
  • Don’t let this bias your conclusions
• Look for trends!