Multi-spectral image labeling with radiometric attribute vectors of image space representation components
View Patent ↗Automatic characterization or categorization of portions of an input multispectral image based on a selected reference multispectral image. Sets (e.g., vectors) of radiometric descriptors of pixels of each component of a hierarchical representation of the input multispectral image can be collectively manipulated to obtain a set of radiometric descriptors for the component. Each component can be labeled as a (e.g., relatively) positive or negative instance of at least one reference multispectral image (e.g., mining materials, crops, etc.) through a comparison of the set of radiometric descriptors of the component and a set of radiometric descriptors for the reference multispectral image. Pixels may be labeled (e.g., via color, pattern, etc.) as positive or negative instances of the land use or type of the reference multispectral image in a resultant image based on components within which the pixels are found.
1. A method for use in classifying areas of interest in overhead imagery, comprising:
organizing, using a processor, a plurality of pixels of at least one input multispectral image of a geographic area into a plurality of components of a hierarchical image representation structure;
deriving, using the processor, at least one set of radiometric descriptors for each component of the plurality of components;
obtaining at least one set of radiometric descriptors for a reference multispectral image, wherein pixels of the reference multispectral image identify at least one land use or land type;
determining, using the processor, for said each component of the hierarchical image representation structure, a similarity metric between the set of radiometric descriptors for said each component and the set of radiometric descriptors of the reference multispectral image, wherein the determined similarity metric for said each component indicates a degree to which the pixels of said each component identify the at least one land use or land type of the reference multispectral image.
2. The method of claim 1 , further including:
obtaining a single radiometric descriptor for each pixel that is representative of a plurality of radiometric descriptors of the pixel at a respectively plurality of spectral bands of the input multispectral image, wherein the organizing utilizes the single radiometric descriptors of the plurality of pixels to generate the hierarchical image representation.
3. The method of claim 2 , wherein the obtaining the single radiometric descriptor for each pixel includes:
determining a minimum difference between the radiometric descriptor of the pixel and the radiometric descriptors of adjacent pixels at each spectral band of the spectral bands, wherein the single radiometric descriptor is a largest difference of the minimum differences at each spectral band of the plurality of spectral bands.
4. The method of claim 2 , wherein the obtaining the single radiometric descriptor for each pixel includes:
determining a maximum difference between the radiometric descriptor of the pixel and the radiometric descriptors of adjacent pixels at each spectral band of the spectral bands, wherein the single radiometric descriptor is a smallest difference of the maximum differences at each spectral band of the plurality of spectral bands.
5. The method of claim 2 , wherein the deriving includes, for each pixel of said each component:
obtaining a set of radiometric descriptors for the pixel, wherein each set includes a plurality of entries; and
using the sets of radiometric descriptors for the pixels of said each component to derive the set of radiometric descriptors for said each component.
6. The method of claim 5 , wherein the set of radiometric descriptors for each pixel is a vector of intensities of the pixel at each spectral band of a plurality of spectral bands of the input multispectral image.
7. The method of claim 5 , wherein the set of radiometric descriptors for each pixel is a vector of intensity gradients of the pixel over the plurality of spectral bands, wherein each intensity gradient is the difference between the intensity of the pixel at one spectral band and the intensity of the pixel at an adjacent spectral band.
8. The method of claim 5 , wherein the using includes:
obtaining means of each respective entry of the sets of radiometric descriptors of the pixels of said each component to obtain radiometric descriptor mean values for corresponding respective entries of the set of radiometric descriptors for said each component.
9. The method of claim 8 , further including:
obtaining variability metrics of each respective entry of the sets of radiometric descriptors of the pixels of said each component to obtain radiometric descriptor variability metrics for the corresponding respective entries of the set of radiometric descriptors for said each component;
analyzing each entry of each set of radiometric descriptors of the pixels in view of the respective radiometric descriptor variability metric of the corresponding entry of the set of radiometric descriptors for said each component; and
making a decision as to whether or not to remove each entry of each set of radiometric descriptors of the pixels from the obtaining means step based on a result of the analyzing step.
10. The method of claim 9 , wherein the analyzing and making steps respectively include:
determining whether or not each entry of each set of radiometric descriptors of the pixels is above a threshold value of the respective radiometric descriptor variability metric of the corresponding entry of the set of radiometric descriptors for said each component; and
removing an entry of a set of radiometric descriptors of the pixels when the entry is above the threshold value.
11. The method of claim 9 , wherein the variability metrics including standard deviations, a range between extreme entries around the mean, or principal entries making up the mean.
12. The method of claim 1 , wherein the obtaining at least one set of radiometric descriptors for the reference multispectral image includes: equating the set of radiometric descriptors of at least one of the components of the hierarchical image representation of the input multispectral image to the set of radiometric descriptors of the reference multispectral image.
13. The method of claim 1 , wherein the obtaining at least one set of radiometric descriptors for the reference multispectral image includes:
obtaining a set of radiometric descriptors for each pixel of a defined portion of the reference multispectral image, wherein each set includes a plurality of entries; and
deriving the set of radiometric descriptors for the reference multispectral image using the sets of radiometric descriptors for the pixels of the defined portion.
14. The method of claim 13 , wherein the reference multispectral image is a portion of the input multispectral image.
15. The method of claim 14 , wherein the multispectral image is non-overlapping with the input multispectral image.
16. The method of claim 13 , further including:
determining, for each pixel of the defined portion, whether its set of radiometric descriptors deviates from the set of radiometric descriptors of the reference multispectral image by more than a threshold; and
re-deriving the set of radiometric descriptors for the reference multispectral image free of any sets of radiometric descriptors of pixels determined to deviate from the derived set of radiometric descriptors of the reference multispectral image by more than a threshold.
17. The method of claim 1 , wherein the determining the similarity metric includes:
ascertaining at least one of a Euclidean Distance, a Manhattan Distance, a Pearson Correlation, or a Cosine Similarity between the set of radiometric descriptors for said each component and the set of radiometric descriptors of the multispectral reference image, wherein the at least one of the Euclidean Distance, the Manhattan Distance, the Pearson Correlation, or the Cosine Similarity is the similarity metric.
18. The method of claim 1 , further including:
mapping the similarity metrics into a resultant image of the geographic area.
19. The method of claim 18 , wherein the mapping includes:
applying a marking to pixels in the input multispectral image that is indicative of the similarity metric of a component within which the pixel is found.
20. The method of claim 19 , wherein the marking includes at least one of a color, pattern, or shade.
21. The method of claim 18 , wherein the determining the similarity metric further includes:
assigning said each component of the hierarchical image representation structure a first normalized similarity metric score when the similarity metric is above a threshold similarity metric indicating that pixels of said each component identify the at least one land use or land type of the reference multispectral image; and
assigning said each component of the hierarchical image representation structure a second normalized similarity metric score when the similarity metric is below a threshold similarity metric indicating that pixels of said each component do not identify the at least one land use or land type of the reference multispectral image.
22. The method of claim 1 , wherein the reference multispectral image is a first reference multispectral image, wherein the pixels of the reference multispectral image identify at least one first land use or land type, wherein the similarity metrics are first similarity metrics, and wherein the method further includes:
obtaining at least one set of radiometric descriptors for a second reference multispectral image, wherein pixels of the reference multispectral image identify at least one second land use or land type, and wherein the first land use or land type is different than the second land use or land type;
determining, using the processor, for said each component of the hierarchical image representation structure, a second similarity metric between the set of radiometric descriptors for said each component and the set of radiometric descriptors of the second reference multispectral image, wherein the determined second similarity metric for said each component indicates indicate a degree to which the pixels of each component identify the at least one second land use or land type; and
mapping the first and second similarity metrics into a resultant image of the geographic area.
23. The method of claim 22 , wherein the mapping includes:
applying a first marking to at least some of the pixels in the input multispectral image that is indicative of the first similarity metric of a component within which the pixel is found; and
applying a second marking to at least some of the pixels in the input multispectral image that is indicative of the second similarity metric of a component within which the pixel is found, wherein the first and second markings are different.
24. The method of claim 23 , wherein the first and second markings include first and second different colors, and wherein a relative intensity of each of the first and second markings indicates a degree to which a respective pixels indicates the first or second land use or land type.