Methods and systems that normalize images, generate quantitative enhancement maps, and generate synthetically enhanced images

Abstract

The current document is directed to digital-image-normalization methods and systems that generate accurate intensity mappings between the intensities in two digital images. The intensity mapping generated from two digital images is used to normalize or adjust the intensities in one image in order to produce a pair of normalized digital images to which various types of change-detection methodologies can be applied in order to extract differential data. Accurate intensity mappings facilitate accurate and robust normalization of sets of multiple digital images which, in turn, facilitates many additional types of operations carried out on sets of multiple normalized digital images, including change detection, quantitative enhancement, synthetic enhancement, and additional types of digital-image processing, including processing to remove artifacts and noise from digital images.

1 . A method for normalizing a pair of images, the method comprising:
 receiving a pair of images;
 registering the pair of images to generate a pair of registered images; 
 normalizing the pair of registered images by
 selecting one image of the pair of registered images as a reference image and the other image of the pair of registered images as a mapped-to image; 
 for each observed intensity value in the reference image,
 selecting a most likely mapped intensity value from the mapped-to image corresponding to the observed intensity value; 
 
 generating an intensity-mapping model from the most likely mapped intensity values; and 
 using the intensity-mapping model to generate a normalized reference image corresponding to the mapped-to image. 
 
   
     
     
         2 . The method of  claim 1  wherein selecting a most likely mapped intensity value from the mapped-to image corresponding to the observed intensity value in the reference image further comprises one of:
 selecting a most likely intensity value from a distribution of mapped-to-image intensity values; and 
 selecting, as the most likely mapped intensity value, an intensity value generated by the mapping function, to which the observed intensity value is input as an argument. 
 
     
     
         3 . The method of  claim 2  further comprising generating a numerical indication of how closely the intensity-mapping function fits the intensity data in the pair of images by:
 summing the absolute values of a set of intensity differences, each intensity difference associated with a first pixel in the reference image and a second pixel in the mapped-to image corresponding to the first pixel and generated by subtracting, from the intensity of the second pixel, a computed intensity value obtained by inputting the intensity of the first pixel to the mapping function to generate the computed intensity. 
 
     
     
         4 . The method of  claim 1  further comprising generating a quantitatively enhanced image from a first-in-time image and a later-in-time image by:
 normalizing the first-in-time-image/later-in-time-image pair; 
 generating a quantitatively enhanced image in which each quantitatively-enhanced-image pixel intensity is obtained from the intensities of the corresponding pixels of the first-in-time image and the later-in-time image by
 computing a difference between the intensity of the corresponding pixel of the later-in-time image and the intensity of the corresponding pixel of the first-in-time image; 
 when the absolute value of the computed difference is less than a threshold value, using 0 as the quantitatively-enhanced-image pixel intensity, and 
 otherwise using the computed difference as the quantitatively-enhanced-image pixel intensity. 
 
 
     
     
         5 . The method of  claim 4  further comprising generating a synthetically enhanced image from the later-in-time image and the quantitatively enhanced image by:
 systematically changing the pixel intensities in the quantitatively enhanced image; and 
 adding the quantitatively enhanced image with systematically-changed pixel intensities to the later-in-time image. 
 
     
     
         6 . The method of  claim 4  further comprising generating an enhanced image sequence by:
 for each successive pair of images in a time-ordered sequence of images,
 generating a quantitatively enhanced image, and 
 storing the quantitatively enhanced image in association with the pair of images. 
 
 
     
     
         7 . The method of  claim 6  further comprising:
 generating a synthetic image for each successive pair of images in a time-ordered sequence of images; and 
 storing the synthetic image in association with the pair of images in a time-ordered sequence of images. 
 
     
     
         8 . The method of  claim 7   wherein, when the time difference between the second and third images in a successive triple of images in the time-ordered sequence of images is greater than a threshold value,
 selecting an intermediate time, 
 generating a synthetic intermediate-time-projected additional image from the quantitatively enhanced image stored in association with the second image, and 
 using, instead of the first image of the triple, the synthetic intermediate-time-projected additional image along with the third image of the triple to generate the quantitatively enhanced image for storage in association with the third image of the triple. 
   
     
     
         9 . The method of  claim 5  wherein systematically changing the pixel intensities further comprises one of:
 linearly scaling the pixel intensities; 
 non-linearly scaling the pixel intensities; and 
 mapping pixel intensities within one or more ranges of pixel intensities to a mapped value. 
 
     
     
         10 . The method of  claim 5  further comprising:
 prior to systematically changing the pixel intensities in the quantitatively enhanced image, denoising the quantitatively enhanced image using a kernel-based denoising method. 
 
     
     
         11 . The method of  claim 10  wherein the kernel-based denoising method comprises:
 using a first noise kernel to identify at least a first-threshold number of noise pixels in the quantitatively enhanced image with absolute-value intensities less than a second-threshold intensity; 
 determining an intensity distribution for the identified noise pixels; 
 using a set of feature-detection kernels to identify texture-associated pixels in the quantitatively enhanced image; 
 using a second cluster-detection kernel to detect clusters of pixels in the quantitatively enhanced image in which all pixel intensities have a common sign; 
 reclassifying non-texture-associated pixels to texture-associated pixels when the cluster in which they are contained has a noise distribution with more than a threshold likelihood of differing from the determined intensity distribution for the identified noise pixels; and 
 setting the intensities of non-texture-associated pixels in the quantitatively enhanced image to 0. 
 
     
     
         12 . The method of  claim 1   wherein each image of the pair of received images, pair of registered images, and normalized reference image is one of
 a two-dimensional image that includes, as image elements, pixels, and 
 a three-dimensional image that includes, as image elements, voxels; 
   wherein each image of the pair of received images, pair of registered images, and normalized reference image is a digitally-encoded image comprising multiple image elements;   wherein each image element includes an encoded value representing an intensity;   wherein each image element is indexable with respect to a coordinate system;   wherein each image element is associated with multiple indices; and   wherein the multiple indices associated with an image element specify a position of the image element within the image.   
     
     
         13 . An image-normalization system comprising:
 a computer system that includes one or more processors, one or memories, and one or more mass-storage devices; and   computer instructions, stored in the one or more memories, that, when executed on the one or more processors, control the computer system to receive a pair of images;   register the pair of images to generate a pair of registered images; and
 normalize the pair of registered images by 
 selecting one image of the pair of registered images as a reference image and the other image of the pair of registered images as a mapped-to image; 
   generating an intensity-mapping model that maps intensities in the reference image to intensities in the mapped-to image; and
 using the intensity-mapping model to generate a normalized reference image corresponding to the mapped-to image; 
   store the normalized images in one or more of the one or more memories and one or more mass-storage devices; and   retrieve the normalized images from the one or more of the one or more memories and one or more mass-storage devices for processing by one or more computational entities to generate additional processed images, including synthetic images and qualitatively enhanced images.   
     
     
         14 . The image-normalization system of  claim 1   wherein each image of the pair of received images, pair of registered images, and normalized reference image is one of
 a two-dimensional image that includes, as image elements, pixels, and
 a three-dimensional image that includes, as image elements, voxels; 
 
   wherein each image of the pair of received images, pair of registered images, and normalized reference image is a digitally-encoded image comprising multiple image elements;   wherein each image element includes an encoded value representing an intensity;   wherein each image element is indexable with respect to a coordinate system;   wherein each image element is associated with multiple indices; and   wherein the multiple indices associated with an image element specify a position of the image element within the image.   
     
     
         15 . The image-normalization system of  claim 13  wherein the intensity-mapping model is a hybrid model that includes both intensity mappings calculated by application of a polynomial mapping model and intensity mappings obtained directly from mapped-to-image intensities. 
     
     
         16 . The image-normalization system of  claim 13  wherein the intensity-mapping model comprises one or more mapping functions, parameters for which are determined using genetic parameter fitting. 
     
     
         17 . The image-normalization system of  claim 13  wherein generating an intensity-mapping model that maps intensities in the reference image to intensities in the mapped-to image further comprises:
 for each observed intensity value in the reference image,
 selecting a most likely mapped intensity value in the mapped-to image corresponding to the observed intensity value; 
 
 generating the intensity-mapping model from the most likely mapped intensity values. 
 
     
     
         18 . The image-normalization system of  claim 17  wherein a quantitatively enhanced image is generated from a first-in-time image and a later-in-time image of a normalized pair of images by:
 generating a quantitatively enhanced image in which each quantitatively-enhanced-image pixel intensity is obtained from the intensities of the corresponding pixels of the first-in-time image and the later-in-time image by
 computing a difference between the intensity of the corresponding pixel of the later-in-time image and the intensity of the corresponding pixel of the first-in-time image; 
 when the absolute value of the computed difference is less than a threshold value, using 0 as the quantitatively-enhanced-image pixel intensity, and 
 otherwise using the computed difference as the quantitatively-enhanced-image pixel intensity. 
 
 
     
     
         19 . The image-normalization system of  claim 18  wherein a synthetically enhanced image is generated from the later-in-time image and the quantitatively enhanced image by:
 systematically changing the pixel intensities in the quantitatively enhanced image; and 
 adding the quantitatively enhanced image with systematically-changed pixel intensities to the later-in-time image. 
 
     
     
         20 . The image-normalization system of  claim 19  wherein an enhanced image sequence is generated by:
 for each successive pair of images in a time-ordered sequence of images, 
 generating a quantitatively enhanced image, and 
 storing the quantitatively enhanced image in association with the pair of images.