Feedback for Part I: Scenes - Dichromatic Reflectance Model & Spatial Statistics

[sytchang]

[1] Usefulness (Part I: Scenes, Section 5): I appreciated the discussion on the dichromatic reflectance model (DRM) and its relevance to computer vision applications for identifying specular reflections. It was interesting to learn how this surface reflectance model can be used to approximate light that is reflected from inhomogeneous dielectric materials but also has its limitations for accurately modeling complex surfaces. DRM describes the reflected light’s color by representing it as a linear combination of the light’s color due to surface reflection and body reflection. DRM is important for color image segmentation as well as photometric invariant feature derivation. Studies have shown that DRM is useful for specular highlight separation and can be used in conjunction with face and skin attributes which constrain color estimates to provide specular highlight removal in facial images. Bright, reflective spots can lead to distortion of key facial features and therefore hinder the performance of recognition systems that are sensitive to changes in appearance. Therefore, the relevance of DRM principles and its role in applications for specular highlight removal are especially crucial for facial recognition, which is used in devices such as smartphones for multi-factor authentication.

[2] Suggested Improvement (Part I: Scenes, Section 6): One suggestion would be to provide specific examples or reference case studies from literature on how spatial statistics are used in the characterization of different natural images. The textbook also mentions that there have been recent developments of machine learning algorithms and diffusion methods which are advancements enabled by the knowledge of spatial statistics. It would be interesting to provide a few examples on how different algorithms may depend on underlying assumptions related to spatial regularites. It would also be useful to reference an ISETCam tutorial script that studies the spatial regularites of images (example: simulation of how camera optics can influence the spatial statistics of different images).