Image Matting with Bayesian and Knockout Matting

This project demonstrates the use of Bayesian Matting and Knockout Matting, techniques used to separate the foreground and background in images.

Dependencies

• numpy
• cv2
• matplotlib

Code Explanation

The main functions in this project are bayesian_matting(compA, compB, backA, backB) and knockout_matting(compA, compB, backA, backB). These functions take in two composite images (compA and compB) and their corresponding background images (backA and backB), and return the estimated foreground and alpha matte.

The functions first resize all images to have the same dimensions. They then calculate per-pixel foreground F and background B color estimates. The background color is adjusted to lie on the line between F and B. The alpha matte, which represents the transparency of the foreground, is then calculated and clipped to the range [0, 1]. Finally, the estimated foreground color is calculated.

Usage

To use this code, you need to have two composite images and their corresponding background images. In this example, we use 'full_blue.jpg', 'full_violet.jpg', 'blue.jpg', and 'violet.jpg'. These images are read in, linearized, and then passed to the bayesian_matting and knockout_matting functions. The estimated foreground and alpha matte are then displayed.

full_blue = plt.imread('full_blue.jpg')
full_violet = plt.imread('full_violet.jpg')
blue = plt.imread('blue.jpg')
violet = plt.imread('violet.jpg')

linear_full_blue = linearize(full_blue)
linear_full_violet = linearize(full_violet)
linear_blue = linearize(blue)
linear_violet = linearize(violet)

# Perform Bayesian Matting
estimated_foreground_bayesian, estimated_alpha_bayesian = bayesian_matting(linear_full_blue, linear_full_violet, linear_blue, linear_violet)

# Perform Knockout Matting
estimated_foreground_knockout, estimated_alpha_knockout = knockout_matting(linear_full_blue, linear_full_violet, linear_blue, linear_violet)

# Display the estimated foreground and alpha matte for Bayesian Matting
plt.imshow(cv2.cvtColor(estimated_foreground_bayesian, cv2.COLOR_BGR2RGB))
plt.title('Estimated Foreground (Bayesian Matting)')
plt.axis('off')
plt.show()

plt.imshow(estimated_alpha_bayesian, cmap='gray')
plt.title('Estimated Alpha Matte (Bayesian Matting)')
plt.axis('off')
plt.show()

# Display the estimated foreground and alpha matte for Knockout Matting
plt.imshow(cv2.cvtColor(estimated_foreground_knockout, cv2.COLOR_BGR2RGB))
plt.title('Estimated Foreground (Knockout Matting)')
plt.axis('off')
plt.show()

plt.imshow(estimated_alpha_knockout, cmap='gray')
plt.title('Estimated Alpha Matte (Knockout Matting)')
plt.axis('off')
plt.show()