Anomalous pixel detection
Systems and methods are disclosed herein to detect pixels exhibiting anomalous behavior in captured image frames. In some examples, temporal anomalous behavior may be identified, such as flickering pixels exhibiting large magnitude changes in pixel values that vary rapidly from frame-to-frame. In some examples, spatial anomalous behavior may be identified, such as pixels exhibiting values that deviate from an expected linear response in comparison with other neighbor pixels.
1. A method comprising:
receiving first and second image frames comprising a plurality of pixels, wherein the first and second image frames are thermal image frames;
selecting kernels of the first and second image frames, wherein each kernel comprises a center pixel and a plurality of neighbor pixels;
comparing a frame-to-frame change of the center pixels with frame-to-frame changes of the neighbor pixels; and
selectively detecting at least one of the center pixels as a temporally anomalous flickering pixel based on the comparing.
2. The method of claim 1 , wherein the comparing comprises comparing a normalized difference of the center pixels to a mean variance of the neighbor pixels.
3. The method of claim 2 , further comprising calculating the normalized difference using the frame-to-frame change of the center pixels and the frame-to-frame changes of the neighbor pixels.
4. The method of claim 2 , further comprising calculating the mean variance from a plurality of variance values associated with the neighbor pixels.
5. The method of claim 4 , further comprising calculating the variance values using the frame-to-frame changes of the neighbor pixels.
6. The method of claim 1 , wherein the kernels are grids of three pixels by three pixels.
7. The method of claim 1 , further comprising assigning a replacement value to the center pixel of at least one of the image frames in response to the detecting.
8. The method of claim 1 , further comprising repeating the selecting, comparing, and detecting for a plurality of different kernels of the first and second image frames to consider a corresponding plurality of center pixels for temporally anomalous flickering pixel behavior.
9. A method comprising:
receiving an image frame comprising a plurality of pixels;
selecting a kernel of the image frame, wherein the kernel comprises a center pixel and a plurality of neighbor pixels;
determining a plurality of linearity measurements based on the center pixel and the neighbor pixels;
updating a spatial anomaly score based on the linearity measurements; and
selectively detecting the center pixel as a spatially anomalous pixel by comparing the spatial anomaly score to a spatial anomaly score threshold.
10. The method of claim 9 , wherein the determining comprises:
calculating a plurality of estimated values for the center pixel using a plurality of corresponding subsets of the neighbor pixels; and
comparing the estimated values to one or more linearity threshold values to determine the linearity measurements.
11. The method of claim 9 , wherein the determining comprises:
identifying a plurality of vectors each comprising the center pixel and a corresponding subset of the neighbor pixels;
for each vector, calculating an estimated value of the center pixel using the corresponding subset of the neighbor pixels; and
comparing the estimated values to one or more linearity threshold values to determine the linearity measurements.
12. The method of claim 9 , further comprising repeating the method for a plurality of image frames to update the spatial anomaly score in response to each image frame.
13. The method of claim 9 , wherein the kernels are grids of five pixels by five pixels.
14. The method of claim 9 , further comprising assigning a replacement value to the center pixel in response to the detecting.
15. The method of claim 9 , further comprising repeating the selecting, determining, and detecting for a plurality of different kernels of the image frame to consider a corresponding plurality of center pixels of the image frame for spatially anomalous pixel behavior.
16. The method of claim 9 , wherein the image frame is a thermal image frame.
17. A method comprising:
receiving first and second image frames comprising a plurality of pixels;
selecting kernels of the first and second image frames, wherein each kernel comprises a center pixel and a plurality of neighbor pixels;
comparing a frame-to-frame change of the center pixels with frame-to-frame changes of the neighbor pixels, wherein the comparing comprises comparing a normalized difference of the center pixels to a mean variance of the neighbor pixels; and
selectively detecting at least one of the center pixels as a temporally anomalous flickering pixel based on the comparing.
18. The method of claim 17 , further comprising calculating the normalized difference using the frame-to-frame change of the center pixels and the frame-to-frame changes of the neighbor pixels.
19. The method of claim 17 , further comprising calculating the mean variance from a plurality of variance values associated with the neighbor pixels.
20. The method of claim 19 , further comprising calculating the variance values using the frame-to-frame changes of the neighbor pixels.