Segmenting the Unseen: Exploring Chromaticity and Parametric Segmentation

Our journey through image processing has led us to tackle more complex challenges, such as segmenting images based on less obvious color details like skin color. Our previous explorations into segmentation using grayscale, RGB, and HSV color channels have paved the way for this understanding. In this sixth reflection, we dive into Chromaticity Segmentation, Parametric Segmentation, Non-parametric Segmentation, and Image Differencing.

Chromaticity Segmentation

The RG Chromaticity space forms the basis of Chromaticity Segmentation. It’s a two-dimensional representation of color that removes the intensity value, focusing instead on the proportions between different color channels mapped in normalized RGB space. This makes it an effective tool for segmenting complex color attributes like skin tones. To compute the RG Chromaticity of an image, we use the following equations:

Since they sum to 1, plotting only red and green should suffice

Parametric Segmentation

With our Chromaticity image in hand, we then proceed to Parametric Segmentation. This technique fits a Gaussian probability distribution to the Chromaticity mask. This distribution helps determine which pixels belong to our color of interest. For this, we need to calculate the mean and standard deviation of our object of interest.

Backs filtered through the Red channel

Bags filtered through the Green channel

Product of the two marks to reveal the orange bag.

Non-Parametric Segmentation

In situations where our region of interest doesn’t align well with a 2D Gaussian function, we resort to a Non-parametric Segmentation method. Here, we utilize the 2D histogram of our reference image, performing histogram backprojection to mask our original image with the computed histogram of our reference patch.

Image Differencing

Finally, we looked into Image Differencing, a simple but effective technique for identifying changes between images. This technique subtracts one image from another and analyses the resulting difference to identify changes or anomalies, which could be very useful in applications like motion detection or anomaly detection.

Image differencing can be good for “Find the differences” games

diff = image1_gray - image2_gray
imshow(diff)

Conclusion

This reflection revealed a whole new dimension in image segmentation, extending our capabilities to discern and highlight specific and even subtle details in images. With Chromaticity Segmentation, Parametric Segmentation, Non-parametric Segmentation, and Image Differencing, we have added powerful tools to our image processing toolbox, empowering us to tackle more intricate and complex image analysis challenges.