High order fractal feature extraction for classification of objects in images

Claim 1

A method for discriminating objects such as targets from non-targets or background in a raw analog image which is pre-processed by being digitized and normalized and consisting of pixels, each of which having its intensity level valued between 0 and 255 where zero is black and 255 is white, and then subjected to a detector capable of identifying possible objects of interest based on appropriate characteristics such as size and brightness for further processing, and providing the x and y center coordinates of each such object in said image, said method having filenames, arrays, and parameters, and said method comprising the steps of: (a) initializing all filenames, arrays, and parameters; (b) inputting normalized image data of pixel intensities; (c) entering input variables consisting of the sizes of a large fractal box, a small fractal box and predetermined threshold test levels; (d) entering the x and y center coordinates of each object detected in said image in the fractal feature array; (e) calculating Sdim, the small box fractal dimension, Bdim, the big box fractal dimension and Fdif, the magnitude of the dimensional differences of Bdim and Sdim for each detected object center; and (f) subjecting said calculated fractal data for each detected object in each said image to classification thresholding where: (1) minimum thresholds for object acceptance using a counter TARGET1; and (2) thresholds for target classification using a counter TARGET.

Claim 2

A method for discriminating objects such as targets from non-targets or background in a raw analog image which is pre-processed by being digitized and normalized and consisting of pixels, each of which having its intensity level valued between 0 and 255 where zero is black and 255 is white, and then subjected to a detector capable of identifying possible objects of interest based on appropriate characteristics such as size and brightness for further processing, and providing the x and y center coordinates of each such object in said image, said method having filenames, arrays, and parameters, and said method comprising the steps of: (a) initializing all filenames, arrays, and parameters; (b) inputting normalized image data of pixel intensities; (c) entering input variables consisting of the sizes of a large fractal box, a small fractal box and predetermined threshold test levels; (d) entering the x and y center coordinates of each object detected in said image in the fractal feature array; (e) calculating Sdim, the small box fractal dimension, Bdim, the big box fractal dimension and Fdif, the magnitude of the dimensional differences of Bdim and Sdim for each detected object center; and (f) subjecting said calculated fractal data for each detected object in each said image to classification thresholding where: (1) minimum thresholds for object acceptance using a counter TARGET1; and (2) thresholds for target classification using a counter TARGET.

Claim 11

A method for discriminating objects such as targets from non-targets or background in a raw analog image which is pre-processed by being digitized and normalized and consisting of pixels, each of which having its intensity level valued between 0 and 255 where zero is black and 255 is white, and then subjected to a detector capable of identifying possible objects of interest based on appropriate characteristics such as size and brightness for further processing, and providing the x and y center coordinates of each such object in said image, said method having filenames, arrays, and parameters, and said method comprising the steps of: (a) initializing all filenames, arrays, and parameters; (b) inputting normalized image data; (c) entering input variables consisting of size of a large and size of a small fractal box and predetermined threshold levels for the input parameters, as defined in Table 4 hereinabove and as identified below comprising: (1) SIGNDIFF, (2) MINSDIM, (3) MAXSDIM, (4) MINBDIM, (5) MAXBDIM, (6) MINFDIF, (7) MINSLOPE, (8) LESSMINSLOPE, (9) MAXPERCENT, (10) MAXINTEN, (11) MINTSUM, (12) MIDTSUM, (13) MAXTSUM, (14) MINDSUM, (15) MINTARGET, (16) MINTARGET2, (17) MAXTARGET1. (d) entering the x and y center coordinates of each object detected in said image in the fractal feature array; (e) calculating Sdim, the small box fractal dimension, Bdim, the big box fractal dimension and Fdif, the magnitude of the dimensional differences of Bdim and Sdim for each detected object center; (f) testing Sdim, the small box fractal dimension, to a threshold level target test for a predetermined type of target MINSDIM≦Sdim≦MSDIM where MINSDIM is the small box fractal minimum threshold and MAXSDIM is the small box fractal maximum threshold, such that if true, TARGET is incremented by 1; (g) performing the fractal difference threshold target test such that if MINSDIM≦Sdim≦MAXSDIM and Fdif≧SIGNDIFF are true, where Fdif is the fractal dimension difference, the value of TARGET is incremented by 1 to establish that an object of some kind and not just background is present; (h) applying a minimum threshold test to look for the minimum fractal feature values to detect an object qualifying as a possible target such that, if Sdim>MINBDIM or Fdif≧MINFDIF are false, where MINBDIM is the big box fractal minimum threshold, the detected object is classified as a non-target and the confidence variable Conf is set to zero and another detected object center is selected for processing, but the test is true the variable TARGET1 is incremented by one and the variable Perpix, the percentage of pixels within the small fractal box above the pixel intensity threshold, is calculated; (i) applying an intensity threshold to the variable Perpix such that if Perpix<MAXPERCENT is true, where MAXPERCENT is the maximum percentage of pixels allowed greater than maximum allowable pixel intensity threshold MAXINTEN, the value of TARGET1 is incremented by one meaning that the object in the defined area is probably a target and not image clutter; (j) calculating fractal slope Slope, the average slope of small box fractal dimension in the vicinity of the detected object; (k) performing a target slope test Slope≦MINSLOPE, the small box fractal dimension minimum slope, to determine if value of Slope is both negative and steep enough to qualify object as possible target and if true incrementing TARGET by one; (l) performing minimum slope test Slope≦LESSMINSLOPE, the small box fractal dimension less minimum slope, to tentatively qualify the object as a target, if true and, incrementing TARGET1 by one, if not true, classifying object as non-target, setting confidence level variable Conf to zero, and going to next detected object center coordinates; (m) performing the minimum score test TARGET≧MINTARGET≧2 and TARGET1≧MINTARGET1≧3 if minimum slope test is true, such that, if the result is false, the detected object is classified as a non-target, the confidence level variable Conf is set to zero, and the next detected object center coordinates are selected for processing, or if the result is true, distribution sums of the small box fractal dimension distribution sum Tsum and the fractal difference sum Dsum are calculated; (n) calculating the small box fractal dimension distribution sum Tsum and the fractal difference distribution sum Dsum for an area around the detected object center to determine target sustainability using the offset matrix of numbered pixels ______________________________________ 71 72 73 74 75 46 47 48 49 50 60 70 45 35 21 22 23 24 25 30 40 65 69 44 34 11 12 13 14 15 29 39 64 68 43 33 1 2 3 4 5 28 38 63 67 42 32 6 7 8 9 10 27 37 62 66 41 31 16 17 18 19 20 26 36 61 51 52 53 54 55 56 57 58 59 76 77 78 79 80 ______________________________________ and starting at pixel #1 and sequentially progressing through pixel #80 or until after Sdim, the small box fractal dimension, and Fdif, the fractal dimension difference, are calculated as centered on a pixel, pass their threshold tests MINSDIM≦SDIM≦MAXSDIM and MINBDIM<Bdim≦MAXBdim and Fdif≧SIGNDIFF, and applying a first Tsum Dsum threshold test Tsum≧MINTsum and Dsum≧MINDSUM are true so that processing of pixels ceases, a tentative target is declared, TARGET1 is incremented by one, and processing proceeds to target classification; and (o) applying a second Tsum, small box fractal dimension distribution sum, and Dsum, big box and small box fractal dimension difference distribution sum, threehold test where said first threshold test is false indicating the detected object may be a weak target, where in said second test said Tsum is greater than or equal to a median value Tsum≧MIDTSUM, the small box fractal dimension distribution sum median threshold, and Dsum is less than a minimum Dsum<MINVDSUM, and where if the result is false, the process proceeds to target classification, and where is the result is true Tsum and Dsum, distribution calculations are performed looping through pixels #1 through #40 using the offset matrix of numbered pixels ______________________________________ 11 7 3 30 10 6 2 o o o o o 16 17 18 38 28 9 5 1 o o o o o o 14 15 36 27 o o o o o o o o o o o 34 24 o o o o o o o o o o o 32 23 o o o o Xc,Yc o o o o o o 31 25 o o o o o o o o o o o 33 26 o o o o o o o o o o o 35 29 12 o o o o o o o o o 19 37 40 13 8 4 o o o o o 21 22 20 39 ______________________________________

Claim 12

A method for discriminating objects such as targets from non-targets or background in a raw analog image which is pre-processed by being digitized and normalized and consisting of pixels, each of which having its intensity level valued between 0 and 255 where zero is black and 255 is white, and then subjected to a detector capable of identifying possible objects of interest based on appropriate characteristics such as size and brightness for further processing, and providing the x and y center coordinates of each such object in said image, said method having filenames, arrays, and parameters, and said method comprising the steps of: (a) initializing all filenames, arrays, and parameters; (b) inputting normalized image data; (c) entering input variables consisting of size of a large and size of a small fractal box and predetermined threshold levels for the input parameters, as defined in Table 4 hereinabove and as identified below comprising: (1) SIGNDIFF, (2) MINSDIM, (3) MAXSDIM, (4) MINBDIM, (5) MAXBDIM, (6) MINFDIF, (7) MINSLOPE, (8) LESSMINSLOPE, (9) MAXPERCENT, (10) MAXINTEN, (11) MINTSUM, (12) MIDTSUM, (13) MAXTSUM, (14) MINDSUM, (15) MINTARGET, (16) MINTARGET2, (17) MAXTARGET1. (d) entering the x and y center coordinates of each object detected in said image in the fractal feature array; (e) calculating Sdim, the small box fractal dimension, Bdim, the big box fractal dimension and Fdif, the magnitude of the dimensional differences of Bdim and Sdim for each detected object center; (f) testing Sdim, the small box fractal dimension, to a threshold level target test for a predetermined type of target MINSDIM≦Sdim≦MSDIM where MINSDIM is the small box fractal minimum threshold and MAXSDIM is the small box fractal maximum threshold, such that if true, TARGET is incremented by 1; (g) performing the fractal difference threshold target test such that if MINSDIM≦Sdim≦MAXSDIM and Fdif≧SIGNDIFF are true, where Fdif is the fractal dimension difference, the value of TARGET is incremented by 1 to establish that an object of some kind and not just background is present; (h) applying a minimum threshold test to look for the minimum fractal feature values to detect an object qualifying as a possible target such that, if Sdim>MINBDIM or Fdif≧MINFDIF are false, where MINBDIM is the big box fractal minimum threshold, the detected object is classified as a non-target and the confidence variable Conf is set to zero and another detected object center is selected for processing, but the test is true the variable TARGET1 is incremented by one and the variable Perpix, the percentage of pixels within the small fractal box above the pixel intensity threshold, is calculated; (i) applying an intensity threshold to the variable Perpix such that if Perpix<MAXPERCENT is true, where MAXPERCENT is the maximum percentage of pixels allowed greater than maximum allowable pixel intensity threshold MAXINTEN, the value of TARGET1 is incremented by one meaning that the object in the defined area is probably a target and not image clutter; (j) calculating fractal slope Slope, the average slope of small box fractal dimension in the vicinity of the detected object; (k) performing a target slope test Slope≦MINSLOPE, the small box fractal dimension minimum slope, to determine if value of Slope is both negative and steep enough to qualify object as possible target and if true incrementing TARGET by one; (l) performing minimum slope test Slope≦LESSMINSLOPE, the small box fractal dimension less minimum slope, to tentatively qualify the object as a target, if true and, incrementing TARGET1 by one, if not true, classifying object as non-target, setting confidence level variable Conf to zero, and going to next detected object center coordinates; (m) performing the minimum score test TARGET≧MINTARGET≧2 and TARGET1≧MINTARGET1≧3 if minimum slope test is true, such that, if the result is false, the detected object is classified as a non-target, the confidence level variable Conf is set to zero, and the next detected object center coordinates are selected for processing, or if the result is true, distribution sums of the small box fractal dimension distribution sum Tsum and the fractal difference sum Dsum are calculated; (n) calculating the small box fractal dimension distribution sum Tsum and the fractal difference distribution sum Dsum for an area around the detected object center to determine target sustainability using the offset matrix of numbered pixels ______________________________________ 71 72 73 74 75 46 47 48 49 50 60 70 45 35 21 22 23 24 25 30 40 65 69 44 34 11 12 13 14 15 29 39 64 68 43 33 1 2 3 4 5 28 38 63 67 42 32 6 7 8 9 10 27 37 62 66 41 31 16 17 18 19 20 26 36 61 51 52 53 54 55 56 57 58 59 76 77 78 79 80 ______________________________________ and starting at pixel #1 and sequentially progressing through pixel #80 or until after Sdim, the small box fractal dimension, and Fdif, the fractal dimension difference, are calculated as centered on a pixel, pass their threshold tests MINSDIM≦SDIM≦MAXSDIM and MINBDIM<Bdim≦MAXBdim and Fdif≧SIGNDIFF, and applying a first Tsum Dsum threshold test Tsum≧MINTsum and Dsum≧MINDSUM are true so that processing of pixels ceases, a tentative target is declared, TARGET1 is incremented by one, and processing proceeds to target classification; and (o) applying a second Tsum, small box fractal dimension distribution sum, and Dsum, big box and small box fractal dimension difference distribution sum, threehold test where said first threshold test is false indicating the detected object may be a weak target, where in said second test said Tsum is greater than or equal to a median value Tsum≧MIDTSUM, the small box fractal dimension distribution sum median threshold, and Dsum is less than a minimum Dsum<MINVDSUM, and where if the result is false, the process proceeds to target classification, and where is the result is true Tsum and Dsum, distribution calculations are performed looping through pixels #1 through #40 using the offset matrix of numbered pixels ______________________________________ 11 7 3 30 10 6 2 o o o o o 16 17 18 38 28 9 5 1 o o o o o o 14 15 36 27 o o o o o o o o o o o 34 24 o o o o o o o o o o o 32 23 o o o o Xc,Yc o o o o o o 31 25 o o o o o o o o o o o 33 26 o o o o o o o o o o o 35 29 12 o o o o o o o o o 19 37 40 13 8 4 o o o o o 21 22 20 39 ______________________________________