Superpixel segmentation methods and systems
Systems and methods for implementing a superpixel image segmentation technique using a boundary preserving distance metric are disclosed. The disclosed technique segments a digital image into superpixels comprising contiguous pixel regions sharing similar characteristics. Superpixel image segmentation techniques presented herein utilize a boundary preserving distance metric. A boundary preserving distance metric presented herein measures the similarity between two pixels of a digital image at least partially based on a boundary probability values of the two pixels and surrounding pixels.
1. A computer-implemented image analysis method of generating a superpixel segmentation of a digitized image, the method to be performed by a system comprising at least one processor and at least one memory, the method comprising:
determining by the system a probabilistic boundary map of the digitized image;
calculating by the system a boundary preserving distance metric between at least one of a plurality of pixels of the digitized image and a neighboring plurality of pixels surrounding the at least one of the plurality of pixels;
segmenting by the system the plurality of pixels based on the calculated boundary preserving distance metric;
generating by the system a plurality of superpixels based on the segmented plurality of pixels; and
outputting by the system the generated superpixels for performing additional digital analysis.
2. The computer-implemented image analysis method of claim 1 , further comprising:
performing by the system a connected component analysis of the segments of the digitized image using the boundary preserving distance metric; and
merging by the system neighboring segments based on results of the connected component analysis.
3. The computer-implemented image analysis method of claim 2 , wherein performing a connected component analysis further comprises:
computing a composite boundary preserving distance metric between a candidate segment and at least one perimeter point in at least one neighboring segment,
wherein the composite boundary preserving distance metric is a minimum of the boundary preserving distance metrics computed between at least one pixel of the candidate segment and the at least one perimeter point, and
wherein the neighboring segment to be merged contains a maximum number of pixels having a minimum composite boundary preserving distance metric; and
wherein merging neighboring segments further comprises:
determining whether a strong boundary exists between the candidate segment and the neighboring segment to be merged; and
merging the candidate segment and the neighboring segment to be merged if no strong boundary exists between the candidate segment and the neighboring segment to be merged.
4. The computer-implemented image analysis method of claim 1 , wherein the boundary preserving distance metric is determined based on a Euclidean distance, a Bhattacharyya distance, and a boundary crossing probability between two pixels.
5. The computer-implemented image analysis method of claim 4 , further comprising measuring the Bhattarachya distance between two pixels based on an adaptive region size determination.
6. The computer-implemented image analysis method of claim 1 , wherein segmenting the plurality of pixels further comprises using a quick shift image segmentation algorithm with the boundary preserving distance metric.
7. The computer-implemented image analysis method of claim 1 , wherein segmenting the plurality of pixels further comprises:
smoothing the boundaries of the segmented plurality of pixels; and
splitting the segmented plurality of pixels.
8. A system for generating a superpixel segmentation of a digitized image comprising:
a memory having program instructions and data storage space;
a processor configured to use the program instructions to perform the steps of:
determining a probabilistic boundary map of the digitized image;
calculating a boundary preserving distance metric between at least one of a plurality of pixels of the digitized image and a neighboring plurality of pixels surrounding the at least one of the plurality of pixels;
segmenting the plurality of pixels based on the calculated boundary preserving distance metric;
generating a plurality of superpixels based on the segmented plurality of pixels; and
outputting the generated superpixels for performing additional digital analysis.
9. The system of claim 8 , wherein the processor is further configured to use program instructions to perform the steps of:
performing a connected component analysis of the segments of the digitized image using the boundary preserving distance metric; and
merging neighboring segments based on results of the connected component analysis.
10. The system of claim 9 , wherein the performing a connected component analysis further comprises:
computing a composite boundary preserving distance metric between a candidate segment and at least one perimeter point in at least one neighboring segment,
wherein the composite boundary preserving distance metric is a minimum of the boundary preserving distance metrics computed between at least one pixel of the candidate segment and the at least one perimeter point, and
wherein the neighboring segment to be merged contains a maximum number of pixels having a minimum composite boundary preserving distance metric; and
wherein merging neighboring segments further comprises:
determining whether a strong boundary exists between the candidate segment and the neighboring segment to be merged; and
merging the candidate segment and the neighboring segment to be merged if no strong boundary exists between the candidate segment and the neighboring segment to be merged.
11. The system of claim 8 , wherein the boundary preserving distance metric is determined based on a Euclidean distance, a Bhattacharyya distance, and a boundary crossing probability between two pixels.
12. The system of claim 11 , wherein the processor is further configured to use the program instructions to perform the step measuring the Bhattarachya distance between two pixels based on an adaptive region size determination.
13. The system of claim 8 , wherein segmenting the plurality of pixels further comprises using a quick shift image segmentation algorithm with the boundary preserving distance metric.
14. The system of claim 8 , wherein segmenting the plurality of pixels further comprises:
smoothing the boundaries of the segmented plurality of pixels; and
splitting the segmented plurality of pixels.
15. A computer-implemented image analysis method of generating a superpixel segmentation of a digitized image, the method to be performed by a system comprising at least one processor and at least one memory, the method comprising:
determining by the system a probabilistic boundary map of the digitized image;
calculating by the system a boundary preserving distance metric between at least one of a plurality of pixels of the digitized image and a neighboring plurality of pixels surrounding the at least one of the plurality of pixels;
segmenting by the system the plurality of pixels based on the calculated boundary preserving distance metric;
performing by the system a connected component analysis of the segments of the digitized image using the boundary preserving distance metric, wherein performing a connected component analysis further comprises: computing a composite boundary preserving distance metric between a candidate segment and at least one perimeter point in at least one neighboring segment, wherein the composite boundary preserving distance metric is a minimum of the boundary preserving distance metrics computed between at least one pixel of the candidate segment and the at least one perimeter point;
generating by the system a plurality of superpixels based on the segmented plurality of pixels; and
outputting by the system the generated superpixels for performing additional digital analysis.