초록 ▼

A computer-implemented method and system for use in alignment of multiple digital images to form a mosaic image involves selecting multiple search site points (SSPs) in an overlapping area of a pair of the digital images and searching for an interesting point (IP) near each of the SSPs. The system i

A computer-implemented method and system for use in alignment of multiple digital images to form a mosaic image involves selecting multiple search site points (SSPs) in an overlapping area of a pair of the digital images and searching for an interesting point (IP) near each of the SSPs. The system involves the calculation of a numeric interest measure (IM) at each of multiple IP-candidate sites near the SSPs. The IM is indicative of the presence of image features at the IP-candidate site and provides a basis for comparing the IP-candidate sites at each of the SSPs and selecting the IP-candidate site having the most significant IM. In a preferred embodiment, IP-candidate sites having an IM that does not exceed a predetermined minimum threshold are discarded. The method also involves locating a tie point (TP) on an overlapping one of the digital images correlating to the IP. The TP together with the IP comprise a tie point pair (TPP) that can be used to calculate and apply geometric transformations to align the images and thereby form a seamless mosaic. The system and method may also involve radiometric balancing of the images to reduce tonal mismatch.

대표청구항 ▼

A computer-implemented method and system for use in alignment of multiple digital images to form a mosaic image involves selecting multiple search site points (SSPs) in an overlapping area of a pair of the digital images and searching for an interesting point (IP) near each of the SSPs. The system i

A computer-implemented method and system for use in alignment of multiple digital images to form a mosaic image involves selecting multiple search site points (SSPs) in an overlapping area of a pair of the digital images and searching for an interesting point (IP) near each of the SSPs. The system involves the calculation of a numeric interest measure (IM) at each of multiple IP-candidate sites near the SSPs. The IM is indicative of the presence of image features at the IP-candidate site and provides a basis for comparing the IP-candidate sites at each of the SSPs and selecting the IP-candidate site having the most significant IM. In a preferred embodiment, IP-candidate sites having an IM that does not exceed a predetermined minimum threshold are discarded. The method also involves locating a tie point (TP) on an overlapping one of the digital images correlating to the IP. The TP together with the IP comprise a tie point pair (TPP) that can be used to calculate and apply geometric transformations to align the images and thereby form a seamless mosaic. The system and method may also involve radiometric balancing of the images to reduce tonal mismatch. od a malignancy of the nodule based on an output of the artificial neural network classifier. 7. The method of claim 6, wherein said determining a likelihood of malignancy step comprises: comparing the output of the artificial neural network classifier with a predetermined threshold; and comparing the output of the artificial neural network classifier with a predetermined threshold; and determining the nodule to be non-malignant when the output of the artificial neural network classifier is less than said predetermined threshold. 8. The method of claim 1, further comprising: extracting features of the nodule based on the outline; applying features including the extracted features to a linear discriminant analysis classifier; and determining a likelihood of malignancy of the nodule based on an output of the linear discriminant analysis classifier. 9. The method of claim 8, wherein said determining a likelihood of malignancy step comprises: comparing the output of the linear discriminant analysis classifier with a predetermined threshold; and determining the nodule to be non-malignant when the output of the linear discriminant analysis classifier is less than said predetermined threshold. 10. The method of claim 1, further comprising: extracting features of the nodule based on the outline; applying features including the extracted features to an artificial neural network classifier and to a linear discriminant analysis classifier; and determining a likelihood of malignancy of the nodule based on outputs of the artificial neural network classifier and the linear discriminant analysis classifier. 11. The method of claim 10, wherein said determining a likelihood of malignancy step comprises: comparing the outputs of the artificial neural network classifier and the linear discriminant analysis classifier with respective predetermined thresholds; and determining the nodule to be non-malignant when both the outputs of the artificial neural network classifier and the linear discriminant analysis classifier are less than said respective predetermined thresholds. 12. The method of claims 6, 7, 8, 9, 10, or 11, wherein: the step of obtaining a digital image comprises obtaining a digital chest radiographic image including the nodule; and the applying step comprises applying plural of the following features: (1) age, (2) root-mean square variation of the power spectrum of the nodule contour, (3) overlap measure in the background trend and the density-corrected image derived from the original image, (4) FWHM for the outside region of the segmented nodule on the background trend and the density-corrected image derived from the original image, (5) degree of irregularity of the nodule outline, (6) FWHM for the inside region of the segmented nodule on the background trend and the density-corrected image derived from the original image, (7) FWHM for the inside region of the segmented nodule on the original image, (8) contrast of the segmented nodule on the background trend and the density-corrected image derived from the original image, (9) contrast of the segmented nodule on the original image, (10) degree of circularity of the nodule outline, (11) relative standard deviation for outside region of the segmented nodule on the background trend and the density-corrected image derived from the original image, and (12) mean pixel value for inside region of the segmented nodule on the background trend and the density-corrected image derived from the original image. 13. The method of claim 12, wherein: the applying step comprises applying each of the following features: (1) age, (2) root-mean square variation of the power spectrum of the nodule contour, (3) overlap measure in the background trend and the density-corrected image derived from the original image, (4) FWHM for the outside region of the segmented nodule on the background trend and the density-corrected image derived from the original image, (5) de