Texture enhancement

The correspondence linking builds a controlled chain of correspondences that can be used for texture enhancement as well. At each reference pixel one may collect a sorted list of image color values from the corresponding image positions. This allows to enhance the original texture in many ways by accessing the color statistics. Some features that are derived naturally from the linking algorithm are:

1. Highlight and reflection removal: A median or robust mean of the corresponding texture values is computed to discard imaging artifacts like sensor noise, specular reflections and highlights[125].

2. Super-resolution texture: The correspondence linking is not restricted to pixel-resolution, since each sub-pixel-position in the reference image can be used to start a correspondence chain. The correspondence values are queried from the disparity map through interpolation. The object is viewed by many cameras of limited pixel resolution, but each image pixel grid will in general be slightly displaced. This can be exploited to create super-resolution texture by fusing all images on a finer resampling grid[77].

3. Best view selection for highest texture resolution: For each surface region around a pixel the image which has the highest possible texture resolution is selected, based on the object distance and viewing angle. The composite image takes the highest possible resolution from all images into account.

An example of highlight removal is shown in Figure 8.3.

Figure 8.3: close-up view (left), 4x zoomed original region (top-right), generation of median-filtered super-resolution texture (bottom-right).