Addressing blur

[soutobias]

What:

Blurring refers to the reduction in image sharpness and detail due to out-of-focus effects, camera motion, or inadequate shutter speed calibration. It can also occur when extracting images from deinterlaced videos. The effect results in a loss of distinctness between pixels, making important features less discernible. Common causes of blurring include camera shake, focus issues, and subject movement. To address blurring, especially without prior knowledge of the blur's cause, blind image deblurring techniques are used [ZHA].

The degradation model for a blurry image can be represented by the equation:

[ S = H \ast U + N ]

where:

• ( S ) is the blurry image,
• ( H ) is the blur kernel (point spread function, PSF),
• ( U ) is the original clear image,
• ( N ) represents noise.

A common approximation for out-of-focus distortion is a circular PSF characterized by its radius ( R ).

Why:

Blurring reduces image resolution and can hinder feature detection, impacting the precision of biodiversity assessments and other image-based analyses.

How:

Blurring can be diagnosed using algorithms to understand how blur spreads within an image set and to set a blur threshold for removing blurred images.

Several techniques can restore a blurred image by estimating the original image from the degraded version:

1. Wiener Filter: This method is used to de-blur an image by reducing the effects of noise and blur, assuming known parameters such as the signal-to-noise ratio (SNR) and the characteristics of the blur kernel.

2. Blind Image Deblurring: Techniques that estimate the blur kernel and use it to restore the original image without prior knowledge of the blur specifics.

Python Code Examples:

1. Blur Detection:

   import cv2
   import numpy as np
   
   def detect_blur(image_path, threshold=100):
       image = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
       laplacian = cv2.Laplacian(image, cv2.CV_64F)
       variance = laplacian.var()
   
       if variance < threshold:
           return True  # Image is likely blurred
       return False  # Image is sharp
   
   # Example usage:
   is_blurry = detect_blur('example_image.jpg')
   print(f"Image is {'blurry' if is_blurry else 'sharp'}")

2. Wiener Filter for De-blurring:

   import cv2
   import numpy as np
   from scipy.signal import convolve2d
   
   def wiener_filter(image, psf, noise_var, estimated_noise_var):
       psf_fft = np.fft.fft2(psf, s=image.shape)
       image_fft = np.fft.fft2(image)
       psf_fft_conj = np.conj(psf_fft)
       result_fft = (psf_fft_conj / (np.abs(psf_fft) ** 2 + noise_var / estimated_noise_var)) * image_fft
       result = np.fft.ifft2(result_fft).real
       return np.clip(result, 0, 255).astype(np.uint8)
   
   def create_psf(radius, size):
       psf = np.zeros((size, size))
       center = size // 2
       y, x = np.ogrid[:size, :size]
       mask = (x - center) ** 2 + (y - center) ** 2 <= radius ** 2
       psf[mask] = 1
       psf /= psf.sum()
       return psf
   
   # Example usage:
   image = cv2.imread('blurry_image.jpg', cv2.IMREAD_GRAYSCALE)
   psf = create_psf(radius=5, size=image.shape[0])
   restored_image = wiener_filter(image, psf, noise_var=0.1, estimated_noise_var=0.2)
   
   cv2.imshow('Restored Image', restored_image)
   cv2.waitKey(0)
   cv2.destroyAllWindows()

What to expect:

• Improved Image Sharpness: The de-blurring process should reduce the blur to the minimum desired level, resulting in a set of images with enhanced clarity and more defined details.

• Blurred image removal: The blurring detection should help refining the image set to the minimum desired level of blur, by image removal.

What makes it difficult:

• Trade-off Between Sharpness and Noise: De-blurring can sharpen images but may also introduce or amplify noise, particularly in low-light conditions. Balancing clarity with noise levels is crucial.

• Introduction of Artifacts: Deblurring techniques may introduce artifacts, which can complicate the analysis and interpretation of the restored images.

Success Metrics:

• Sharpness and Clarity: The success of de-blurring is measured by the improved sharpness and clarity of the restored images. Effective de-blurring should result in images with well-defined details and reduced blur, while maintaining a balance between enhancement and noise.

[soutobias]

@Mojtabamsd , take a look on the develop branch. I perform some changes in the blur method. I also apply several changes in the code, mainly related on how we call the functions and in the configuration file.