Method for detecting pixel outliers
A method for identifying singleton outlier pixels in a selected color space in a digital image including a plurality of pixels, includes for each 3×3 patch of pixels in the image, calculating the diameter of the 3×3 patch of pixels. For each pixel in the patch, the distance to its nearest neighbor pixel within the patch is computed, as measured in the selected color space. The computed distance from each pixel in the patch is compared to its nearest neighbor with a threshold that is a preselected fraction of the diameter. A center pixel in the patch is identified as an outlier pixel if its calculated distance to its nearest neighbor is the largest distance to a nearest neighbor and exceeds the threshold.
1. In a digital image including a plurality of pixels, a method for identifying singleton outlier pixels in a selected color space comprising:
for each 3×3 patch of pixels in the image, calculating the diameter of the 3×3 patch of pixels;
for each pixel in the patch, computing the distance to its nearest neighbor pixel within the patch, as measured in the selected color space;
comparing the computed distance from each pixel in the patch to its nearest neighbor with a threshold that is a preselected fraction of the diameter; and
identifying a center pixel in the patch as an outlier pixel if its calculated distance to its nearest neighbor is the largest distance to a nearest neighbor and exceeds the threshold.
2. The method of claim 1 wherein the selected color space is sRGB space.
3. The method of claim 1 wherein the distance is the squared Euclidean distance.
4. The method of claim 1 , wherein the preselected fraction of the diameter is about 0.1.
5. In a digital image including a plurality of pixels, a method for identifying pairs of adjacent outlier pixels in a selected color space comprising:
for each 3×3 patch of pixels in the image, calculating the minimum diameter of the 3×3 patch of pixels;
for each pixel in the 3×3 patch, computing the distance to its nearest neighbor pixel within the 3×3 patch, as measured in the selected color space;
comparing the computed distance from each pixel in the 3×3 patch to its nearest neighbor with a threshold that is a preselected fraction of the minimum diameter;
if the calculated distance to the nearest neighbor of a center pixel in the 3×3 patch is not both the largest distance to a nearest neighbor in the 3×3 patch and does not exceed the threshold, making a set of comparisons of the computed distance from non-center pixels in the 3×3 patch to its nearest neighbor with a threshold that is a preselected fraction of the diameter, each comparison in the set made by ignoring a different one of the non-center pixels in the 3×3 patch;
if in only one of the comparisons in the set, the calculated distance from the center pixel to its nearest neighbor is the largest distance to a nearest neighbor and exceeds the threshold:
define a new 3×3 patch of pixels having the ignored pixel in the identified comparison as the new center pixel;
for each pixel in the new 3×3 patch, computing the distance to its nearest neighbor pixel within the new 3×3 patch, as measured in the selected color space;
comparing the computed distance from each pixel in the new 3×3 patch to its nearest neighbor with a threshold that is a preselected fraction of the minimum diameter;
if the calculated distance from the new center pixel to its nearest neighbor in the new 3×3 patch of pixels is the largest distance to a nearest neighbor and exceeds the threshold, identify the center pixel and the new center pixel as a pair of adjacent outlier pixels.
6. The method of claim 5 wherein the selected color space is sRGB space.
7. The method of claim 5 wherein the distance is the squared Euclidean distance.
8. The method of claim 5 , wherein the preselected fraction of the minimum diameter is about 0.1.