JP4544891B2 - Image processing method and program for line extraction, line concentration image filter - Google Patents

Description

  The present invention relates to an image processing method for line extraction capable of extracting only line information from an image at high speed, a program for executing the same, and a line for performing image processing for line extraction by mounting the same on a computer. The present invention relates to a concentration image filter.

  When analyzing an object in an image, lines that specify the shape of the image are important. One of these lines is a boundary that separates the object and the background, and the other is a normal meaning line having an elongated width on the image, that is, a line segment. The latter is often an image of a long and slender object that rises like a kamaboko, and in some cases an image showing a recessed groove or the like.

  By the way, the boundary on the image can be detected by detecting a portion where a large change in luminance appears by taking the differentiation or difference because the line originally has no width and the shadow of the object and the background has contrast. . Further, since a line (line-shaped convex region) indicating a raised elongated object appears with a width on the image, a method using a second-order differential of a plane in consideration of contrast and the line width (Gaussian Laplacian) Recognition is performed by extracting luminance ridge line information by using a method for estimating the ridge from the curvature and the like. However, since these methods are affected by contrast, line width, and noise, they greatly depend on the image to be analyzed.

  For example, medical images such as X-ray photographs and CT images have a lot of medical information, but the contrast between the target region, the surrounding region, and the background is generally dark and diverse. In addition, since a three-dimensional shape, a structure such as a blood vessel, a fiber, and a nerve is displayed two-dimensionally, it is difficult to extract a lesion from a shadow. An example is a medical image of a tumor. When there is a tumor shadow in the image, it is necessary to distinguish between a malignant tumor and a benign tumor. In many cases, a malignant tumor is called a spicule with various contrasts, widths and patterns extending radially from the center of the tumor. A white line is generated. However, experience is required to recognize this spicula, and in order to extract it from a medical image, it is necessary to cut off the influence of unnecessary shadows. When trying to extract a spicule using the above-mentioned second-order differentiation, pre-processing and threshold setting are required to remove the influence of contrast. The calculation for performing such processing is complicated and time consuming, and the result is noisy.

Therefore, a method based on a line concentrated vector field model has been proposed (see, for example, Non-Patent Documents 1 and 2). This is a line-convergence vector field that detects line information using a model of a linear region without being affected by contrast, line width, and the like. In this model, the intensity of the first-order differential vector (hereinafter referred to as luminance gradient vector) is ignored, and only the direction distribution of the vector is focused. In the case of an ideal linear convex region, the ridge line has the maximum luminance value. At this time, the luminance gradient vector has a direction orthogonal to the ridge line, is distributed to face both sides of the ridge line, and is concentrated toward the ridge line. Hereinafter, such a ridge line on which the luminance gradient vector is concentrated is referred to as a vector convergence line. FIG. 12A is an explanatory diagram of the definition of the line concentration degree, and FIG. 12B is an explanatory diagram of the line concentration degree in an infinitely long line concentration vector field. When the brightness gradient vector of a point P of the vector concentration line V _C vicinity is linear concentrations, the luminance gradient vector is orthogonal to the vector concentration line V _C, it becomes dotted vector of FIG. 12 (a). However, the actual luminance gradient vector g _P is often inclined, and when the angle θ _P between the two is represented, it represents the degree to which the angle θ _P is concentrated. Therefore, to define a line concentration degree C _P as the evaluation value of the linear concentrations vector field for this vector concentrate line V _C as follows. That is, C _P = cos θ _P (g _P ≠ 0) and C _P = 0 (g _P = 0). Also defines the linear region A including the neighboring point P, a line concentration degree C _A and the average value of the line concentration degree C _P of all points P in the region. Therefore, if the vector concentration line V _C is a true line, C _A = 1.

In Non-Patent Document 1, the position, direction, and scale of the line concentrated vector field are estimated as follows. Attention point P (x, y) assumes a vector concentration line V _C having a street predetermined inclination φ of the assumed distance w away linearly parallel to the vector concentration line V _C as a search line S _w. The average value of the degree of line concentration of each point Q on the search line S _w and the degree of concentration C _S (w) are obtained. At this time, C _S (w) = E [cos θ _Q ] (QεS _w ). Here, E [] indicates an average. Since linear concentrations vector field is the region that has spread to both sides of the vector concentration line V _C, assuming the area is A, maximum search width is W, the distance assumptions from exploration line S _w until both area end W _r , W _l , the line concentration degree C _{A with} respect to the vector concentration line V _C in the area _A is the average of the line concentration degrees of the search lines S _w in the area, and C _A = E [C _S (w)] (S _w ΕA). When the attention point P (x, y) passes through the true vector concentration line V _C and the inclination φ and the distances W _r and W _l coincide with the true values, C _A becomes the maximum value. Therefore, _C angle which the value becomes maximum of _A phi, distance _W r, to explore the _{W l,} obtains a vector concentration line _{V C} at the point of interest P (x, y), the line of the maximum value of this _{C A} The degree of concentration may be C (x, y). This line concentration degree C (x, y) is referred to as a full width adaptive region line convergence degree (hereinafter referred to as FA line concentration degree). This is expressed as C (x, y) = max _φ [max _Wr {max _Wl C _A }] (W _r > 0, W _l > 0). In this way, the line concentration vector field is determined.

Incidentally, in the linear concentrations vector model shown in FIG. 12 (b), target point P (x, y) assumes a vector concentration line V _C on adaptive linear concentration vector field area A is assumed to around , FA concentration C (x, y) = max _φ [{max _Wr C _Ar + max _Wl C _Al } / 2] is obtained, and the line widths W _r and W _l are estimated. Here, C _Ar and C _Al are the average of the concentration degrees of the luminance gradient vectors in the region estimated in the r direction and the l direction. Linear concentrations vector field with respect to the line-shaped protruding region showing the elongated object raised, vector concentrate line V _C is present in the center, it becomes an infinite length in the width 2W.

The vector concentration line V _C measurement point at a distance w from Q (x, y) when calculating the degree of concentration, the search line S _w from l direction (on the left) with V _C from l to measurement point Q is present In the region up to the boundary of the direction, all gradient vectors are in the r direction (rightward in the drawing), the same as the direction with respect to V _C , and the degree of concentration C _S (w) = 1. In the region from the search line S _w to V _C , the direction of l is the opposite direction to V _C , so the degree of line concentration C _S (w) = − 1, and in the r direction from the search line S _w to V _C. It is the same as the direction, and the concentration degree C _S (w) = 1.

Therefore, _{C Ar} = 1 in the region of the distance _{W r} from the search line _{S w.} On the other hand, if the width of W _l is different in the l direction, C _Al changes. That _{0 <W l <C Al =} -1 when the _w, w _<W l _<C when the _{w + W Al = (W l} -2w) / W l, when the _{w + W <W l C Al} = (W the -2w) / _{W l.} Since W _l = w + W when W _l coincides with the boundary, C _Al = (W−w) / (W + w) and becomes the maximum, so that the region can be estimated regardless of the position of the measurement point Q. Further, in the entire region, the line concentration degree C (x, y) = {1+ (W−2w) / W} / 2 = ½ + (W−2w) / 2W = 1−w / W. Vector concentrate line _{V C} on a line concentration degree C (x, y) = 1 and also linear concentrations vector field inside C (x, y) is seen to be a 0.5 or higher.

However, the foregoing description is that in the absence of noise, when there is actual noise as the image processing, the method by the conventional FA area line concentration degree becomes difficult detection of vector concentrate line V _C. FIG. 13 is a flowchart for calculating the line concentration of image processing based on the FA concentration based on the GAWHT method (Gradient-Angle-Weighted Hough Transform). As shown in FIG. 13, first, the maximum search width W and length l for determining the search region, the search line direction φ, φ and the luminance gradient vector direction θ are divided into n, the original number. The luminance I (x, y) of the image is input (step 101), and the position of the measurement point Q (x, y) is initialized (step 102). Next, a luminance gradient vector is calculated from the input luminance I (x, y) (step 103), and (i, j) of (i, j) is set assuming that a neighborhood region is set in the search region. The range is calculated (step 104).

Subsequently, the line concentration degree on the evaluation line within the range of (i, j) in the vicinity is calculated (step 105). ρ represents a discretized distance to the evaluation line . These h _cr (ρ) and h _cl (ρ) are Hough transforms of partial straight lines only within the range of (i, j). The h _cr (ρ) and h _cl (ρ) calculated in this way are changed to ρ = 0 to W _{r in the r} direction and ρ = 0 to W _l in the l direction, and h _cr (ρ ), H _cl (ρ), and the maximum value of C _H (x, y), and W _r and W _l at that time are obtained (step 106). Further, line information on the screen is extracted by moving the position of the measurement point Q (x, y) (step 107).

  As can be seen from the above explanation, the method based on the FA area line concentration degree is not affected by the width of the line concentration vector field even if the contrast is uneven, but it requires a huge amount of calculation time and noise. It was easy to mix, and the influence appeared in both the direction φ and the line width.

As in Non-Patent Documents 1 and 2, numerical value using a density gradient vector (using sin θ _P ) is performed, entropy is calculated from the probability density Pi of the histogram, and a reference candidate line having a minimum value of entropy A method for recognizing an irradiation field of a radiographic image in which a contour is determined by searching for the image is also proposed (see Patent Document 1). However, this technique extracts the boundary of the surface and does not extract only the line information of the linear convex region, and does not have a problem of speeding up the extraction of the line information at high speed.

JP-A-10-275213 IEICE Transactions, J81-D-II, No.11, November 1998, IEICE, Yukiyoshi Yoshinaga 1 "Detection of line area not affected by contrast and width using luminance gradient vector" Method ", p. 2547-p. 2555 IEICE Transactions, Vol. 1, No. 1, January 2004, IEICE, Yukiyoshi Yoshinaga, et al., 4 “Spicular Discrimination Method Using Feature Extraction of Lines by Luminance Gradient Vector Field Model” ", P. 146-p. 153

As described above, the vector concentration of the luminance corresponding to the ridge of the line (linear convex region) indicating the elongated object raised in the image using the conventional method of estimating the ridge from the second derivative or the curvature, etc. method of extracting a line V _C is the width of the contrast and the line, due to the influence of noise, the analysis results to the image quality of the image to be analyzed is highly dependent.

  The analysis of the linear region based on the line concentration vector field model of Non-Patent Documents 1 and 2 is not affected by the contrast or the width of the line concentration vector field. However, the image quality is uneven due to the influence of noise. The same applies to the technique of Patent Document 1. And when performing the calculations of Non-Patent Documents 1 and 2, a measurement method based on the conventional GAWHT method is used, but it is necessary to calculate the FA line concentration and the line width by performing the enormous calculation as described above. There was a big problem in practical use. Therefore, when noise is mixed, there is a case where the image quality is poor and the reliability of the analysis result remains a problem even though enormous calculation is performed.

  Therefore, in order to solve such a problem, the present invention provides an image processing method for line extraction that extracts line information in an image without being affected by image contrast, line width, noise, and the like, It is an object of the present invention to provide a program for executing this and a line concentration degree image filter.

The present invention provides a measurement vicinity area having a fixed shape fixed to both wings of a search line passing through each measurement point for a plurality of measurement points on an image, and is included in the vicinity area Measure the direction of the luminance gradient vector of the image at multiple neighboring points, calculate the degree of concentration that evaluates the difference between the direction of the luminance gradient vector and the direction of the search line at each of the neighboring points, and all the degree of concentration in the neighboring region An image processing method for line extraction that calculates a line concentration degree for a measurement point from a line and determines that there is line information along a search line when the line concentration level reaches a maximum. The direction of the search line and the direction of the search line are respectively discretized, and each of the directions based on the direction of the luminance gradient vector, the position of the pixel in the vicinity area, and the number of pixels for each direction of the search line on the local coordinates indicating the vicinity area. Can be shared at measurement points Leave computing the fundamental sum of concentration of the neighboring points, when measured orientation of the intensity gradient vectors, summed in neighboring region to choose one from the candidates of the basic addition value for each direction of the search line and calculates the linear concentrations of each measurement point based on the basic additional value by.

  According to the line concentration image filter and program of the present invention, and the image processing method for line extraction, it is possible to extract line information in an image without being affected by the contrast of the image, the width of the line, noise, and the like. it can.

In order to solve the above-described problems, the first embodiment of the present invention provides a measurement neighborhood having a fixed shape fixed to both wings of a search line passing through each measurement point with respect to a plurality of measurement points on the image. Concentrate to prepare an area, measure the direction of the brightness gradient vector of the image at a plurality of neighboring points included in the neighboring area, and evaluate the difference between the direction of the brightness gradient vector and the direction of the search line at each of the neighboring points A line extraction that calculates the degree of line, calculates the line concentration for the measurement point from all the degrees of concentration in the neighborhood, and determines that there is line information along the search line when the line concentration reaches a maximum Image processing method for discriminating the direction of the luminance gradient vector and the direction of the search line, respectively , and the direction of the luminance gradient vector and the neighboring region for each direction of the search line in advance on the local coordinates indicating the neighboring region The position of the pixel in Leave computing the fundamental sum of the concentration level of the neighboring point that can be shared by each measurement point based on the number of pixels, when measuring the orientation of the intensity gradient vectors, the candidate of the basic addition value for each direction of the search line An image processing method for line extraction, characterized by calculating a line concentration degree at each measurement point based on a basic addition value by selecting one of them and adding them in a neighboring region, The calculation at each measurement point is accelerated and the measurement is performed in the direction of the luminance gradient vector by calculation in the vicinity area having a fixed shape, further separation of the basic addition value that can be shared and accumulation of the preliminary calculation results. At the same time, since it is a neighboring region having a certain shape, line information in the image can be extracted without being affected by the contrast of the image, the line width, noise, or the like.

  The second form of the present invention is a form subordinate to the first form, characterized in that the neighboring region is a fixed rectangular region that is long and long with a predetermined width along both wings of the search line. This is an image processing method for line extraction, which requires a small amount of calculation and can speed up the calculation.

  A third aspect of the present invention is a form subordinate to the first or second form, wherein the neighborhood region is composed of two regions separated by both wings of the search line, respectively. Is an image processing method for noise and has extremely high resistance to noise.

  The fourth form of the present invention is a form dependent on any one of the first to third forms, and the degree of concentration c (t) is c (t) = c (−t), c (t + π) = An image for line extraction, characterized in that c (t), c (0) = 1, c (π / 2) = 0, and calculated by a function indicating monotonic decrease in 0 ≦ t ≦ π This is a processing method, and a basic addition value that can be shared can be calculated in advance.

  A fifth aspect of the present invention is a form that is subordinate to any one of the first to fourth aspects, and is characterized in that it is determined that there is line information when the line concentration degree is greater than 0.5. When the line concentration degree is 0.5 or less, information on the boundary of the surface area is included, and this can be removed together with other noises.

According to a sixth aspect of the present invention, there is provided a computer, (1) a plurality of measurement points on an image, a fixed shape vicinity for measurement on both wings of a search line passing through each measurement point Rutotomoni provided regions, respectively discretized orientation and the search line of intensity gradient vectors of a plurality of neighboring points included in the near-neighbor region, the brightness gradient vector for each direction of the search line in the local coordinates representing the area near Basic addition value calculation means for calculating and storing a basic addition value of the degree of concentration of neighboring points that can be shared by each measurement point based on the orientation of the pixel and the position and number of pixels in the neighboring region , (2) Search line direction Is set, the direction of the brightness gradient vector of the image is measured at each of the neighboring points at each measurement point, and the basic addition value corresponding to the direction is extracted from the candidates for the basic addition value and added in the vicinity region. Especially Ri line concentration degree calculated based on the basic sum value, and outputs the direction of the search line when the line concentration degree and / or that it can be determined that the line information in the line concentration degree showing maximum values at each measurement point line concentration degree calculation means, a program of the order to function as a line extracting unit determines a line information is greater than (3) line concentration degree calculation means line concentration degree predetermined noise immunity value calculated, discrete Calculation in a neighboring area with a fixed shape, separation of basic addition values that can be shared, and accumulation of preliminary calculation results, the calculation at each measurement point is accelerated, and the direction of the luminance gradient vector In addition to measurement, since it is a neighboring region having a certain shape, line information in the image can be extracted without being affected by the contrast of the image, the width of the line, noise, and the like.

  The seventh aspect of the present invention is a program subordinate to the sixth aspect, characterized in that the noise tolerance value is 0.5, and when the line concentration degree is 0.5 or less, Information on the boundary of the surface area is included, but it can be removed along with other noises.

An eighth aspect of the present invention is a program subordinate to the sixth or seventh aspect, wherein the basic addition value calculation means sets a rectangular area as a neighboring area , and the amount of calculation is There are few, and calculation can be sped up.

The ninth form of the present invention is, (1) for a plurality of measurement points on the image, provided the area near the immobilized predetermined shape for measurement to the wings of the search line through the respective measuring points Rutotomoni, The direction of the brightness gradient vector and the direction of the search line of a plurality of neighboring points included in the neighborhood area are discretized, respectively, and the direction of the brightness gradient vector and the direction in the neighborhood area are determined for each direction of the search line on the local coordinates indicating the neighborhood area. Basic addition value calculation means for calculating and storing a basic addition value of the degree of concentration of neighboring points that can be shared by each measurement point based on the position and the number of pixels of (2) , and (2) when the direction of the search line is set , Measure the direction of the brightness gradient vector of the image at each of the neighboring points at each measurement point, extract the basic addition value corresponding to the direction from the candidates for the basic addition value, and add them in the neighborhood area to add the basic addition value line concentration on the basis of Was calculated, and the line concentration degree calculation means for outputting a direction of the search line when the line concentration degree and / or that it can be determined that the line information in the line concentration degree showing maximum values at each measurement point, (3) A line concentration degree image filter characterized by comprising: line extraction means for determining line information when the line concentration degree calculated by the line concentration degree calculation means is greater than a predetermined noise tolerance value, The calculation at each measurement point is accelerated and the measurement is performed in the direction of the luminance gradient vector by calculation in the vicinity area having a fixed shape, further separation of the basic addition value that can be shared and accumulation of the preliminary calculation results. At the same time, since it is a neighboring region having a certain shape, line information in the image can be extracted without being affected by the contrast of the image, the line width, noise, or the like.

  The line extraction image processing according to the first embodiment of the present invention will be described with reference to FIG. 1A is an explanatory diagram of an actual line of an actual image in the first embodiment of the present invention, FIG. 1B is an explanatory diagram of a luminance vector distribution of the actual line of FIG. 1A, and FIG. FIG. 2B is an explanatory diagram of ideal luminance vector distribution with respect to lines in Example 1 of the present invention, and FIG. 2B is an explanatory diagram of luminance vectors in the line area of FIG.

  First, the line extraction of the first embodiment uses a narrow fixed neighborhood around the search line and calculates the concentration level with respect to the search line to obtain a vector concentration line. The principle of this line extraction will be described. .

  When line information is extracted from an image as shown in FIG. 1A, the presence of a line is usually recognized along the ridge line of the linear convex region that appears in the image. If the point of interest P is moved in the image of FIG. 1A and a ridge line can be extracted, a line segment can be extracted from the image. The right diagram in FIG. 2A shows the state of the luminance and the luminance gradient vector of such a linear convex region, and shows the state of an ideal linear convex region vector that appears on the image. According to this, in the ideal state, a uniform density change is formed from the vicinity of the edge of the linear convex region toward the center, and the luminance gradient vectors are all aligned vertically toward the central ridge line. Therefore, in the present invention, when the direction of the luminance gradient vector is orthogonal to the ridge line from both sides of the ridge line, it is determined that the line segment is a ridge line (hereinafter referred to as a vector concentration line V).

The left diagram in FIG. 2A is an ideal luminance gradient vector model distribution for determining a linear convex region. When the attention point P (x, y) exists on the vector concentration line V, the neighboring region r (neighboring region r ₁ ) on one side of the vector concentration line V and the neighboring region r (neighboring region r ₂ ) on the other side. In each case, the luminance gradient vector concentrates vertically toward the vector concentration line V. However, such an ideal neighboring region r rarely appears in an actual image, and actually appears as a neighboring region r as shown in FIG. 1B, and the luminance gradient vector is expressed as in the neighboring region r ₁ . Although the directions are generally opposite to _{each other} in the vicinity region r ₂ , the lines are concentrated separately toward the vector concentration line V. Further, when the attention point P (x, y) is not on the vector concentration line V, the luminance gradient vector is directed in a completely different direction, and is directed in a direction to which irregularities such as noise are added.

Therefore, in the present invention, in order to obtain the vector concentration line V, an assumed vector concentration line V (hereinafter referred to as a search line S) passing through the point of interest P (x, y) is assumed, and a neighboring region r _{1 with} respect to this search line S. And an evaluation value for the attention point P (x, y) at all the neighboring points Q (x, y) in the neighboring region r ₂ , that is, a line concentration degree C _N (x, y) described below, When the line concentration degree C _N (x, y) indicates a predetermined value 1 or the maximum value indicating the vector concentration line V, this direction is set as the line direction. The line concentration degree C _N (x, y) due to this narrow fixed neighborhood region of the present invention is referred to as a narrow width fixed line convergence degree (hereinafter referred to as NF line concentration degree).

As shown in FIG. 2B, if the direction of the vector concentration line V and the search line is φ, the direction of the ideal luminance gradient vector is ψ, and the direction of the actual luminance gradient vector is θ, the neighboring region r ₁ The luminance gradient vector ψ at the neighboring point Q ₁ belonging to and the Q ₂ belonging to the neighboring region r ₂ is orthogonal to the vector concentration line V, and in the neighboring region r ₁ ψ is ψ = φ + π / 2, and in the neighboring region r ₂ ψ Becomes ψ = φ−π / 2. By the way, the actual luminance gradient vector in Q (x, y) is inclined by an angle (θ−ψ) with the direction ψ of the ideal luminance gradient vector due to the irregularity of the vector flow. Therefore, the actual luminance gradient vector is evaluated in relation to the above model using (θ−ψ) as the evaluation index of each luminance gradient vector and the concentration degree c (θ−ψ) as the evaluation value. Note that θ and ψ are functions of x and y. The function c (t) representing the degree of concentration has c (t) = c (−t), c (t + π) = − c (t), c (0) = 1, c (π, where t is a radian value. / 2) = 0, and it may be a monotonously decreasing function with 0 ≦ t ≦ π. In the first embodiment, c (θ−ψ) = cos (θ−ψ) is employed as the degree of concentration c (θ−ψ). Note that the concentration degree c (θ−ψ) is not limited to cos (θ−ψ) as can be seen from the above description.

When the true vector concentration line V faces the direction φ of the line segment, Q (x, y) in the vicinity regions r ₁ and r ₂ of both wings of the search line S in the direction φ passing through P (x, y). ) Of the brightness gradient vector of θ) is θ = ψ, the concentration degree c (θ−ψ) of each point is the maximum, and c (0) = 1 on the vector concentration line V. Conversely, when the degree of concentration c (θ−ψ) shows the maximum value in the neighboring regions r ₁ and r ₂ , the direction of φ is the direction of the vector concentration line V. At this time, the neighboring regions r ₁ and r ₂ The brightness gradient vector of the neighboring point Q (x, y) is a vector orthogonal to the direction φ. When the degree of concentration c (θ−ψ) reaches the maximum value 1 in all the neighboring regions r ₁ and r ₂ , the attention point P (x, y) is on the vector concentration line V, and the vector concentration Line V indicates that it is directed in the direction φ. In practice, even if it is not exactly 1, if it is close to 1, it is generally in the direction of φ.

Therefore, with respect to the assumed target point P (x, y), the neighboring point Q (x _d , y _d ), the width w between the vector concentrated line V and the search line S, the vector concentrated line V, the direction φ of the search line S In addition, assuming near-width fixed regions r ₁ and r ₂ each having a width w _max and a length l in the vicinity of P (x, y), the line concentration degree C _N (x, y) with respect to the search line S Perform the calculation. Assuming that the sum of the areas of the narrow-fixed neighboring regions r ₁ and r ₂ is A (r), the line concentration degree C _N (x, y) can be expressed by (Equation 1). FIG. 3 is an explanatory diagram of line concentration in the vicinity of the fixed narrow width according to the first embodiment of the present invention. W is the maximum search width of the search area , and W ≧ 2w _max . However, the width w _max is better to choose a sufficiently smaller value than the W so as not to affect W is, in the real image it is sufficient to take the _{w max =} 2~5 degree.

図３によれば、近傍領域ｒ_２内における輝度勾配ベクトルの集中度ｃ（θ-ψ）は１、近傍領域ｒ_１内におけるベクトル集中線Ｖと探索線Ｓ間の輝度勾配ベクトルの集中度ｃ（θ-ψ）は−１、近傍領域ｒ_１内の残りの領域の輝度勾配ベクトルの集中度ｃ（θ-ψ）は１であるから、Ｃ_Ｎ（ｘ，ｙ）＝｛（ｗ_ｍａｘ−ｗ）×１＋（−１）×ｗ＋１×ｗ_ｍａｘ｝／２ｗ_ｍａｘ＝１−ｗ／ｗ_ｍａｘとなる。ベクトル集中線Ｖと探索線Ｓが一致したときは、注目点Ｐ（ｘ，ｙ）はベクトル集中線Ｖ上に存在し、Ｃ_Ｎ（ｘ，ｙ）＝１となることが分る。

According to FIG. 3, the concentration degree c (θ−ψ) of the luminance gradient vector in the neighboring region r ₂ is 1, and the concentration degree c of the luminance gradient vector between the vector concentration line V and the search line S in the neighboring region r ₁ . Since (θ−ψ) is −1 and the degree of concentration c (θ−ψ) of the luminance gradient vector in the remaining region r ₁ is 1, C _N (x, y) = {(w _max − w) × 1 + (− 1) × w + 1 × w _max } / 2w _max = 1−w / w _max When the vector concentrated line V and the search line S coincide with each other, it can be seen that the attention point P (x, y) exists on the vector concentrated line V and C _N (x, y) = 1.

As described above, in the image processing according to the first embodiment, the assumed attention point P (x, y) is successively moved with respect to the image to be processed, and each of the adjacent narrow regions r ₁ and r ₂ is assumed. If (Expression 1) is calculated and a search line S with C _N (x, y) = 1 is obtained, it can be seen that this line is a vector concentration line V that is a ridge on the image.

By the way, “cos (θ−φ + π / 2)” in the second term of (Equation 1) can be rewritten as “−cos (θ−φ−π / 2)”. Both terms use the same calculation formula “cos (θ−φ−π / 2)”, and C _N (x, y) is the same calculated value “cos (θ−φ) in the neighboring regions r ₁ and r ₂ . -π / 2) / A (r) "is added or subtracted. The present invention utilizes this property, divides the direction φ and the direction θ of the luminance gradient vector into n, and performs basic addition of a plurality of calculated values “cos (θ−φ−π / 2) / A (r)”. This value is stored in advance as a value. Accordingly, the basic addition value of (Equation 1) is calculated as the first step, and the direction θ of the luminance gradient vector in the narrow fixed regions r ₁ and r ₂ is measured as the second step to obtain the direction φ and From this combination, one corresponding basic addition value is extracted from the candidates accumulated in advance, and is added and subtracted to achieve high speed.

On the other hand, in the conventional FA line concentration method, the range of the neighboring point (i, j) is calculated from the length l of the neighboring regions r ₁ and r ₂ and the maximum search width W, and this neighboring point (i, j) Measurement of the line concentration with respect to the search line S and the width estimation of W _r and W _l must be executed in addition to this, and a huge amount of calculation is required. Addition and subtraction can be performed, and the amount of calculation can be drastically reduced.

Furthermore, in the present invention, it is possible to exclude not only a linear convex region indicating a line but an edge portion indicating a surface, and only a linear convex region can be extracted. In the case of the linear convex region shown in FIG. 3, the concentration degree c (θ−ψ) of the brightness gradient vector is 0 when w> W. On the other hand, when this is a vector field indicating a surface, the right half surface of the vector concentration line V is w> W and the concentration degree c (θ−ψ) is 1. At this time, C _N (x, y) = {(w _max −W) × 1 + (− 1) × W} / 2 w _max = ½−W / w _max , and C _N (x, y) is 0.・ Do not exceed 5. That is, by setting a threshold value of 0.5 to the line concentration degree C _N (x, y) and extracting C _N (x, y)> 0.5, such an edge portion or the like can be excluded from the extracted data. It is.

  Therefore, the image processing procedure for line extraction according to the first embodiment of the present invention will be described below with reference to the flowcharts of FIGS. FIG. 4 is a flowchart of basic addition value calculation of image processing for line extraction in Embodiment 1 of the present invention, and FIG. 5 is a flowchart of calculation of line concentration degree of image processing for line extraction in Embodiment 1 of the present invention. FIG. 6 is an explanatory diagram of a pixel arrangement in a fixed region in the narrow width in the first embodiment of the present invention, FIG. 7 is a flowchart of line extraction processing of image processing for line extraction in the first embodiment of the present invention, and FIG. FIG. 9 is a block diagram of a line concentration filter according to the first embodiment of the invention. FIG. 9 shows the distribution of the line concentration by the FA method, the narrow fixed vicinity region of the first embodiment, and the narrow fixed fixed vicinity region of the second embodiment. FIG. The image processing procedure of the first embodiment includes (1) a basic addition value calculation step, (2) a line concentration degree calculation step, and (3) a line extraction step.

First, (1) the basic addition value calculation step will be described. Assuming narrow areas of fixed widths r ₁ and r ₂ having a width w _max and a length l on both wings of the search line S that passes through the point P (x, y) of the image, the direction φ and the brightness of the search line S are assumed. A division number n for discretizing the direction of the gradient vector is set (step 1). The local coordinates with the first P (x, y) as the origin are prepared, and the positions of the pixels constituting the neighboring regions r ₁ and r ₂ having the width w _max and the length l are set (step 2). The number of pixels included in the neighboring areas r ₁ and r ₂ is counted, and M ₁ and M ₂ are obtained (step 3).

Subsequently, as shown in FIG. 6, sequential numbers m are assigned to the pixels in the neighboring areas r ₁ and r ₂ , and the neighboring point Q (x _d , y _d ) in the neighboring area r ₁ is expressed as a function of this m. position coordinates of _{(d 1 x (m),} d 1 y (m)), the position coordinates showing the vicinity of point Q in the neighboring region _{r 2} as a function of the _{m (d 2 x (m)} , d 2 y (M)) is calculated (step 4). Next, t = (n / 2π) θ is set, φ is discretized with n, and (Equation 2) and (Equation 3) are calculated with the neighboring regions r ₁ and r ₂ (step 5). That is, in the neighborhood region r ₁ (Equation 2), m = 0 to M ₁ , t = 0 to (n−1). Note that φ is discretized for each Δφ .

同様に、近傍領域ｒ_２で（数３）、ｍ＝０〜Ｍ_２、ｔ＝０〜（ｎ−１）となる。

Similarly, in the neighborhood region r ₂ (Equation 3), m = 0 to M ₂ and t = 0 to (n−1).

続いて、近傍領域ｒ_１，ｒ_２の画素の連番を結合する（ｓｔｅｐ６）。すなわち、近傍領域ｒ_１においてＣ（ｔ，ｍ）＝Ｃ_１（ｔ，ｍ）、ｄｘ（ｍ）＝ｄ_１ｘ（ｍ）、ｄｙ（ｍ）＝ｄ_１ｙ（ｍ）とし、ｍ＝０〜Ｍ_１、ｔ＝０〜（ｎ−１）である。また、近傍領域ｒ_２において、Ｃ（ｔ，ｍ＋Ｍ_１）＝Ｃ_２（ｔ，ｍ）、ｄｘ（ｍ＋Ｍ_１）＝ｄ_２ｘ（ｍ）、ｄｙ（ｍ＋Ｍ_１）＝ｄ_２ｙ（ｍ）とし、ｍ＝０〜Ｍ_２、ｔ＝０〜（ｎ−１）とするものである。

Subsequently, the sequential numbers of the pixels in the neighboring regions r ₁ and r ₂ are combined (step 6). That is, C (t, m) = C ₁ (t, m), dx (m) = d ₁ x (m), dy (m) = d ₁ y (m) in the neighborhood region r ₁ , m = 0 ~M _1, is t = 0~ (n-1) . In the neighboring region r ₂ , C (t, m + M ₁ ) = C ₂ (t, m), dx (m + M ₁ ) = d ₂ x (m), dy (m + M ₁ ) = d ₂ y (m) , M = 0 to M ₂ , t = 0 to (n−1).

In this way, m (m = 0 to M) for specifying the position of each pixel, the total number of pixels M = M ₁ + M ₂ , the direction φ of the search line S, and the luminance gradient vector by the basic addition value calculation step of the first embodiment. (2) Basic addition values C ₁ (t, m) and C _{2 that} are candidates for the degree of concentration of each pixel used in the calculation of the line concentration calculation step. (T, m) is expressed as one basic addition value C (t, m), where m = 0 to M, t = 0 to (n−1), calculated in advance and stored in the storage means. be able to. Similarly, dx (m) and dy (m) can be specified by m = 0 to M. The parameter φ is also changed at φ = 0 to (n−1), and the basic addition value C (t, m) is calculated at each φ.

By the way, the degree of concentration c (x) is c (x) = c (−x), c (x + π) = − c (x), c (0) = 1 when the y-axis is taken in the φ direction. , C (π / 2) = 0, and 0 ≦ x ≦ π, which is a monotone decreasing function, and “cos (θ−φ−π / 2)” “cos (θ−φ) in (Expression 1)” Instead of “+ π / 2)”, other functions f (θ−φ−π / 2) and f (θ−φ + π / 2) satisfying this condition can be used. As such a function, there is a power function of a cosine function such as cos ^k (θ−φ−π / 2), where k is an odd number. At this time, since f (θ−φ + π / 2) = − f (θ−φ−π / 2), the calculation of step 5 is C ₁ (t, m) = f (t in the neighboring region r ₁ . _{-φ, d 1 x (m)} , d 1 y (m)), m = 0~M 1, t = 0~ the (n-1), in the region near _{r 2, C 2 (t,} m) = f (t−φ, d ₂ x (m), d ₂ y (m)), m = 0 to M ₂ , and t = 0 to (n−1) may be used. In the case of this cos ^k (θ−φ−π / 2), since it rapidly decreases near θ = φ + π / 2, a highly sensitive line filter can be obtained.

Subsequently, the (2) line concentration degree calculation step executed when calculating the line concentration degree of the actual image will be described. (1) Since the basic addition value that is a candidate for the degree of concentration of each pixel used in the next step is calculated in advance in the basic addition value calculation step, the division for dividing the line direction φ and the luminance gradient vector direction θ If the number n and the luminance I (x, y) of the original image are input, the line concentration degree C _N (x, y) is the sum of the basic addition values C (t, m) in each pixel according to (Formula 1). It is obtained as

In FIG. 5, φ (in other words, the number of divisions n) and the luminance I (x, y) are input (step 11), and in calculating the position of the attention point P (x, y) in the image, the neighboring point Q (x _d, the position coordinates and the line concentration degree _C N (x of _{y d), y)} to initialize (step 12). Thereby, the direction φ of the search line and the coordinates (dx (m), dy (m)) of the neighboring areas r ₁ and r ₂ can be expressed by n and m (m = 0 to M).

Subsequently, the direction θ of the luminance gradient vector of the neighboring point Q is obtained (step 13), t = (n / 2π) θ is calculated, and is converted into an integer (step 14). Further, the presence / absence of φ input is checked (step 15). If φ is not specified, φ = 0 is initialized (step 16). Optionally when there is specification of phi is set as φ = φ ₀ (step17). m is initialized as m = 0 (step 18). _{Then, x = x d + dx (} m), performs calculation of _{y = y d + dy (m} ), calculates the coordinates of the target point P (x, y) (step19 ). Then, one of the candidates for the basic addition value C (t, m) is selected and C _N (x, y) = C _N (x, y) + C (t, m) is calculated (step 20). The line concentration degree C _N (x, y) is obtained.

Next, m = m + 1 is calculated (step 21), and it is determined whether or not m matches M (step 22). If m does not match M, the process returns to step 19 to repeat the calculation of coordinates. If m matches with M in step 22, whether or not φ is specified is checked again (step 23). If φ is not specified, φ = φ + 1 is calculated (step 24), and whether φ matches (n−1). In addition, if φ is specified and φ = φ ₀ , it is determined whether it matches (φ ₀ +1) (step 25). If both do not match, return to step 18 to calculate. repeat. If there is designation of phi = phi ₀ in step 23, or if phi matches the (n-1) at step 25, the neighboring point _{Q (x d, y} d) whether or not the mobile needs of the determination ( Step 26) If the movement is not necessary, the measurement of Q (x _d , y _d ) is terminated. If the movement is necessary, the measurement is moved to a new Q (x _d , y _d ), and the process returns to step 14 and is repeated.

Further, (3) the line extraction step will be described. In this step, when the line concentration degree C _N (x, y) is larger than the noise tolerance value 0.5, it is determined as a linear convex region, and C _N (x, y) A portion that can be evaluated as a line in the current image is extracted according to the value of and is output as line information. It should be noted that in the following line extraction step, enhancement processing is performed in order to output a line image more clearly. In the line extraction process, when the line concentration C _N (x, y) is input as shown in FIG. 6 (step 31), a coordinate system with the attention point P (x, y) in the image as the origin is selected, The position coordinates of the neighboring point Q (x _d , y _d ) are initialized (step 32). A maximum value is obtained from the input line concentration C _N (x, y) for each φ, and whether or not C _N (x, y)> 0.5 is the maximum value of C _N (x, y). Is determined (step 33), and when C _N (x, y)> 0.5, C _L (x, y) = 1 is output (step 34). If there is no designation of φ, φ indicating the maximum value at the same time is output as φ indicating the direction of the line. In step 33, if C _N (x, y) ≦ 0.5, C _L (x, y) = 0 is output (step 35), and it is necessary to move the neighboring point Q (x _d , y _d ). Is determined (step 36), the measurement of Q (x _d , y _d ) is terminated if no movement is necessary, and the process returns to step 33 for a new Q (x _d , y _d ) if movement is necessary. repeat.

Incidentally, the noise tolerance value 0.5 described above is based on the fact that C _N (x, y) does not exceed 0.5 for the line indicated by the edge portion of the surface region as described above. . The data indicating the edge portion can be removed from the line image output by the process of (3) line extraction step. Even if there is other noise, C _N (x, y) is detected only in a very narrow small area of the ridge line in the first embodiment, so that the influence of noise is almost eliminated and only the line image is obtained. It can output reliably.

As described above, in the first embodiment of the present invention, (1) the basic addition value calculation step, the basic addition value of each pixel used in the calculation of the subsequent (2) line concentration calculation step every φ, m, t in advance. C (t, m) is calculated, and in the (2) line concentration calculation step, t is calculated by obtaining the direction θ of the luminance gradient vector of the neighboring point Q, and the basic addition value C (t, m) Is used to calculate the line concentration C _N (x, y) for each known φ ₀ or each φ, and (3) C _N (x, x, which can be evaluated as a linear convex region from this set in the line extraction step. Since y) is calculated, unlike the conventional GAWHT method, the calculation of the line concentration C _N (x, y) and the line widths W _r and W _l is not required to be performed at the same time. By simply selecting the basic addition value C (t, m) obtained by calculation and repeating simple addition, The line-shaped protruding region in the image within the time, in particular can be extracted only that ridge line.

In addition, when a specific φ ₀ is specified and sliced in an object space composed of (x, y, φ), information on this φ ₀ direction is extracted, and further detailed analysis becomes possible. For example, φ is designated as φ = φ _A and φ _B , and C _N (x, y, φ _A ) and C _N (x, y, φ _B ) at this time are obtained, and 0.5 respectively be extracted is greater than, allowing separation of the line information facing the line information and phi _B direction toward the phi _a direction. and space sliced in phi _A, in the space that slice phi _B, C _N without specifying φ _(x, y, φ) information that was eliminated in the line information having the maximum value, extraction with respective spaces It becomes possible.

For example, when two lines in the directions of φ _A and φ _B intersect, a line indicated by an object on the camera side is extracted in one slice space and the lower object is indicated in the other slice space. A line is extracted. In many cases, the lower line is extracted so that there is a defect in the crossing portion. Since the outline of the missing portion is not clear, the output image is reduced to reduce the amount of information in the unclear portion, and φ is changed again to output a line image in the C _N (x, y, φ) space. Etc., the three-dimensional structure can be clarified more clearly.

Furthermore, not only binary image extraction in which line information with C _N (x, y, φ) larger than 0.5 is extracted, but also 0.5 to 0.6, 0.6 to 0.7, 0.7 It is possible to extract multi-value images such as .about.0.8 and 0.8 or more, and it is possible to make the image easier to see by adding gradation to the image.

  As described above, the image processing for line extraction according to the first embodiment of the present invention not only changes the direction φ, but also enables processing of a three-dimensional image by specifying φ and processing. Then, by outputting as an object space of (x, y, φ), various image analysis becomes possible.

  Subsequently, the block configuration of the program and the line concentration filter of the first embodiment that executes the image processing described above will be described. In FIG. 8, 1 is a basic addition value calculation means for executing the procedure of the above-mentioned (1) basic addition value calculation step, 2 is a line concentration degree calculation means for executing the procedure of (2) the line concentration degree calculation step, Reference numeral 3 denotes line extraction means for executing the procedure of (3) line extraction step.

Basic addition value computing means 3 counts the number of pixels of each pixel constituting the neighboring region r _1, r _2, to calculate the position of the neighboring point Q in the neighboring region r _1, r _2, theta, discrete and φ The basic addition values (Equation 2) and (Equation 3) corresponding to all changes in θ and φ are calculated. Each of the basic sum of neighboring regions r _1, r ₂ is, (θ-φ) are those that utilize the property of the same across the region near r _1, r ₂ If so, the neighboring region r _1, r By combining the serial numbers of the _two pixels, the calculation can be performed using the serial numbers. Further, the line concentration degree calculating means 2 calculates the direction θ of the luminance gradient vector, and when φ is known for all neighboring points Q, the basic addition value is simply added from the values of θ and φ to obtain C _N (x, y) is calculated. When φ is not yet determined, the discretized φ is changed, and the basic addition value is added for each φ to calculate C _N (x, y) that is the maximum value. Thus, φ and C _N (x, y) are obtained. Furthermore, the line extraction means 3 outputs only the value of C _N (x, y) calculated by the line concentration degree calculation means 2 that is larger than the noise tolerance value 0.5, and C _N (x, y)> 0.5. _If, C N _(x, y) and _{C L (x, y) =} 1, C N (x, y) when a _{≦ 0.5, C N (x,} y) and _C L (x, y) = As 0, the line image can be sharpened. This is not limited to binary, and can be executed in multiple values.

  The program according to the first embodiment is read into a central processing unit (CPU) of a computer (not shown), and includes (1) a basic addition value calculation step, 2) a line concentration degree calculation step, and (3) a line extraction step. The calculation step is executed according to the procedure. The line image filter according to the first embodiment is such that this program is installed in a computer and processed according to the procedure, and is configured as a function realization means by the computer and the program.

Therefore, a comparison is made between the image processing for line extraction in the vicinity region having the fixed narrow width described above and other processing. 9A shows the conventional FA line concentration degree, and the NF line concentration degree shown in FIG. 9B shows the line concentration degree in the vicinity of the fixed width in the first embodiment. The FA line concentration is 1 at the ridge line, and the line concentration is 0.5 at the line width boundary. Rather than using luminance differentiation, the line concentration is calculated in the direction of the vector, so it is not affected by contrast, but the line concentration is calculated together with the line width estimation in a wide area, so it is affected by noise. easy. On the other hand, the line concentration degree in the vicinity region with the fixed fixed width according to the first embodiment is such that the line concentration degree is small only in a very narrow small region of about unit width, and only in a region where C _N (x, y) exceeds 0.5. It can be seen that only the ridge line can be extracted sharply without being affected by noise.

  Therefore, the result of actually performing the image processing of the first embodiment and other processing on the medical image will be described. FIG. 10 is a photograph obtained by analyzing a mass shadow of breast cancer using the line concentration degree image filter of Example 1 of the present invention and another filter. 10A is an X-ray photograph of the original image, FIG. 10B is a binary image processed by the line image filter of the first embodiment, and FIG. 10C is a multi-value processed by the line concentration degree image filter of the first embodiment. (D) is a binary image processed by the FA line concentration filter, (e) is a multi-value image processed by the FA line concentration filter, and (f) is a binary image processed by a conventional differential filter. .

  The X-ray photograph that is the original image of (a) is shown with a blurred shadow of the tumor, and it is difficult to read the information unless it is an expert. Conventionally, this information is differentiated by a differential filter, and judged as an image as shown in (f). According to (f), it can be seen that the white ring at the center is raised in a lot of noise, and radial lines that appear to be blood vessels and spicules extend around it, but it is very unclear. However, there is a lot of miscellaneous information with contrast, and the white ring shows the boundary of the shadow of the tumor by differentiation, making it difficult to see the spicula.

  On the other hand, in the binary image processed by the line concentration degree image filter of Example 1 in (b), only the line is detected, the ring-shaped part indicating the boundary is removed, and the long and radially extending spicule A linear portion such as a large number of blood vessels is clearly extracted. According to the multi-value image of (c), a clearer and more three-dimensional line structure of the tumor shadow is extracted, and it is easy to distinguish between a malignant tumor and a benign tumor. Extraction of lines is important for diagnosis of malignant tumors, and it is necessary to cut off the influence of noise, but the line concentration degree image filter of Example 1 can be said to be a filter suitable for this purpose. Next, in the binary image processed by the FA line concentration filter of (d), the lines are extracted with the correct thickness, and an excellent result almost the same as the line concentration image filter of Example 1 is shown. However, the lines that are messy and intertwined are conspicuous. This is because the FA line concentration filter is more susceptible to noise than the line concentration image filter of the first embodiment, and the calculation amount and the calculation time are 2 to 2 more than those of the line concentration degree image filter of the first embodiment. 4 digits higher. This is because the line-concentration image filter of the first embodiment uses a neighborhood region in which the width is fixed, uses the characteristics of cosine function for evaluating the degree of concentration, discretization of φ and θ, This is because the calculation of the basic addition value has been successfully separated. Since the basic addition value is stored in advance, the number of times of measurement of θ of the luminance gradient vector is at least 1 / M (M is the total number of pixels). Because.

  As described above, the line concentration degree image filter and the program according to the first embodiment and the image processing method for line extraction extract line information in an image without being affected by the contrast, line width, noise, and the like of the image. be able to. In addition, since only line information can be extracted from an image in the first embodiment, a line image obtained by extracting the line information of the first embodiment and a line image obtained by general differentiation are output from one image, and the difference between the two is output. It is also possible to extract an image of only the boundary portion of the surface area.

A line extraction image processing method and program, and a line concentration degree image filter according to the second embodiment of the present invention will be described with reference to FIG. FIG. 11 is an explanatory diagram of line concentration in the vicinity of the fixed fixed width in the second embodiment of the present invention. Image processing of the second embodiment separates the neighboring region r _1, r ₂ of the wings of the search line S, the search line neighboring region r ₁ from the position apart by a width w _min respectively from S each _direction, r ₂ It is provided. Accordingly, the configuration is basically the same as that of the first embodiment, and a detailed description will be given to the first embodiment, and only differences will be described. 1, 2, and 4 to 10 are also referred to in the second embodiment.

In the second embodiment, as shown in FIG. 11, the points from the attention point P (x, y) on the search line S to the neighboring point Q (x _d , y _d ), the vector concentration line V, and the search line S. The width w, the vector concentrated line V, and the direction φ of the search line S are set, and each of the widths w _max and the length l are separated from a position separated by a width w _{min in} the vicinity of P (x, y). The line concentration C _N (x, y) with respect to the search line S is calculated assuming the neighboring regions r ₁ and r ₂ . When the areas of the V-direction unit lengths of the narrow fixed regions r ₁ and r ₂ are A ₁ (r) and A ₂ (r), A (r) = A ₁ (r) + A ₂ (r). From (Equation 1), the line concentration degree C _N (x, y) can be expressed. In the case of a real image, this w _min may be about w _min = 1 and w _max ≧ 2. It is sufficient that _wmax = 2 to 5 or so. In this case, noise resistance is further improved by the line extraction image processing of the second embodiment as compared to the first embodiment.

Similar to the first embodiment, the image processing procedure for line extraction according to the second embodiment includes (1) a basic addition value calculation step, (2) a line concentration degree calculation step, and (3) a line extraction step. The The second embodiment is basically the same as the first embodiment. However, the coordinates at _{m = M 1 (dx (m} ), dy (m)) is greatly changed.

  As shown in FIG. 8, the program and the line concentration degree image filter of the second embodiment are also (1) basic addition value calculation means 1 for executing the basic addition value calculation step procedure, and (2) the procedure of the line concentration degree calculation step. The line concentration degree calculating means 2 for executing (3), and the line extracting means 3 for executing the procedure of (3) line extraction step.

Basic addition value computing means 3 counts the number of pixels of each pixel constituting the neighboring region r _1, r _2, to calculate the position of the neighboring point Q in the neighboring region r _1, r _2, theta, discrete and φ The basic addition values (Equation 2) and (Equation 3) corresponding to all changes in θ and φ are calculated. Further, the line concentration degree calculating means 2 calculates the direction θ of the luminance gradient vector, and when φ is known for all neighboring points Q in each of the neighboring regions r ₁ and r ₂ , the line concentration degree calculating means 2 calculates the basis from the values of θ and φ. The addition value is simply added to calculate C _N (x, y). When φ is not yet determined, the discretized φ is changed, and the basic addition value is added for each φ to calculate C _N (x, y) that is the maximum value. Thus, φ and C _N (x, y) are obtained. Furthermore, the line extraction means 3 outputs only the value of C _N (x, y) calculated by the line concentration degree calculation means 2 that is larger than the noise tolerance value 0.5, and C _N (x, y)> 0.5. _If, C N _(x, y) and _{C L (x, y) =} 1, C N (x, y) when a _{≦ 0.5, C N (x,} y) and _C L (x, y) = As 0, the line image is sharpened.

A description will be given of the line concentration degree of image processing for line extraction in the vicinity of the narrow fixed area according to the second embodiment. The two-divided NF line concentration shown in (c) of FIG. 9 shows the line concentration in the vicinity of the fixed width in the second embodiment. The degree of line concentration due to the narrow fixed region in Example 2 is a very narrow small region (w _max −w _min ), and more than 0.5 (w _min + w _max ) / 2, which is a slight value of 2 to w _min . It can be seen that only the ridge line can be extracted sharply, with only the area having the influence of noise and being hardly affected.

In the calculation in the first embodiment, the accuracy is lower than the other portions at the end portions of the neighboring regions r ₁ and r ₂ , but in the second embodiment, the accuracy is much less at the end portions of the neighboring regions r ₁ and r _2. There is an excellent characteristic that it does not deteriorate. Further, when the line on the image crosses, the line concentration filter of the second embodiment can detect this with high sensitivity.

  The present invention can be applied to an image filter and image processing capable of extracting line information in an image.

(A) Explanatory drawing of actual line of real image in embodiment 1 of the present invention, (b) Illustrated distribution distribution of luminance vector of actual line of (a) (A) Explanatory diagram of distribution of ideal luminance vector with respect to line in embodiment 1 of the present invention, (b) Illustrated explanatory diagram of luminance vector of line area in (a) Explanatory drawing of the line concentration in the near fixed area | region of Example 1 of this invention Flowchart of calculation of basic addition value of image processing for line extraction in embodiment 1 of the present invention Flowchart of calculating line concentration degree of image processing for line extraction in embodiment 1 of the present invention Pixel array explanatory diagram of a narrow fixed region in the first embodiment of the present invention Flowchart of line extraction processing of image processing for line extraction in Embodiment 1 of the present invention Block configuration diagram of a line concentration filter in Embodiment 1 of the present invention Distribution diagram of the degree of line concentration by the FA method, the narrow fixed neighborhood region of the first embodiment, and the narrow fixed neighborhood region divided and divided by the second embodiment The photograph which analyzed the tumor shadow of the breast cancer with the line concentration degree image filter of Example 1 of this invention and another filter Explanatory drawing of the line concentration in the vicinity area | region of fixed narrowness in Example 2 of this invention (A) Explanatory drawing of definition of line concentration degree (b) Explanatory drawing of line concentration degree in infinite length line concentration vector field Flow chart for calculating the line concentration of image processing based on the FA concentration based on the GAWHT method

Explanation of symbols

1 Basic addition value computing means 2-wire concentration calculating means 3-wire extracting means P (x, y) point of interest near point _Q (x _{d, y} d) neighboring point S search line V concentrate line theta, phi direction r _1, r ₂ , r neighboring region w, w _max , w _min width W maximum search width C _N (x, y) line concentration c (θ−ψ) concentration

Claims (9)

For a plurality of measurement points on the image, a measurement neighborhood area having a fixed shape fixed to both wings of the search line passing through each measurement point is prepared, and a plurality of neighborhood points included in the neighborhood area are prepared. Measuring the direction of the luminance gradient vector of the image, calculating the degree of concentration for evaluating the difference between the direction of the luminance gradient vector and the direction of the search line at each of the neighboring points, and calculating all the degree of concentration in the neighboring region An image processing method for line extraction that calculates a line concentration degree for the measurement point from and determines that there is line information along the search line when the line concentration degree reaches a maximum.
The direction of the brightness gradient vector and the direction of the search line are discretized, respectively, and the direction of the brightness gradient vector and the position of the pixel in the vicinity area for each direction of the search line in advance on local coordinates indicating the vicinity area And calculating a basic addition value of the degree of concentration of the neighboring points that can be shared by each measurement point based on the number of pixels,
When measuring the direction of the luminance gradient vector, each one based on the basic addition value by selecting one of the candidates for the basic addition value for each direction of the search line and adding within the neighborhood region An image processing method for line extraction, wherein the line concentration degree of a measurement point is calculated. 2. The image processing method for line extraction according to claim 1, wherein the neighboring area is a fixed rectangular area that is long and has a predetermined width along both wings of the search line. The image processing method for line extraction according to claim 1, wherein the neighboring area includes two areas separated by both wings of the search line. The degree of concentration c (t) satisfies c (t) = c (−t), c (t + π) = − c (t), c (0) = 1, c (π / 2) = 0, and 0 The image processing method for line extraction according to claim 1, wherein the image processing method is calculated by a function indicating monotonic decrease in ≦ t ≦ π. 5. The image processing method for line extraction according to claim 1, wherein when the line concentration degree is larger than 0.5, it is determined that there is line information. Computer
(1) For a plurality of measurement points on the image, a fixed region for measurement having a fixed shape is provided on both wings of a search line passing through each measurement point, and a plurality of neighborhoods included in the vicinity region The direction of the brightness gradient vector of the point and the direction of the search line are discretized, and the direction of the brightness gradient vector, the position of the pixel in the neighborhood area, and the pixel for each direction of the search line on the local coordinates indicating the neighborhood area A basic addition value calculating means for calculating and storing a basic addition value of the degree of concentration of the neighboring points that can be shared by each measurement point based on the number;
(2) When the direction of the search line is set, the direction of the luminance gradient vector of the image is measured at each of the neighboring points at each measurement point, and the basis corresponding to the direction among the candidates for the basic addition value The line concentration is calculated based on the basic addition value by taking out the addition value and adding it in the vicinity region, and the line concentration can be determined as line information in the line concentration degree indicating the maximum value at each measurement point. A line concentration degree calculating means for outputting the degree and / or the direction of the search line at that time,
(3) A program for functioning as line extraction means for determining line information when the line concentration degree calculated by the line concentration degree calculation means is greater than a predetermined noise tolerance value. The program according to claim 6 , wherein the noise tolerance value is 0.5. The program according to claim 6 or 7, wherein the basic addition value calculation means sets a rectangular area as the neighboring area. (1) For a plurality of measurement points on the image, a fixed region for measurement having a fixed shape is provided on both wings of a search line passing through each measurement point, and a plurality of neighborhoods included in the vicinity region The direction of the brightness gradient vector of the point and the direction of the search line are discretized, and the direction of the brightness gradient vector, the position of the pixel in the neighborhood area, and the pixel for each direction of the search line on the local coordinates indicating the neighborhood area Basic addition value calculation means for calculating and storing a basic addition value of the degree of concentration of the neighboring points that can be shared by each measurement point based on the number;
(2) When the direction of the search line is set, the direction of the luminance gradient vector of the image is measured at each of the neighboring points at each measurement point, and the basis corresponding to the direction among the candidates for the basic addition value The line concentration is calculated based on the basic addition value by taking out the addition value and adding it in the vicinity region, and the line concentration can be determined as line information in the line concentration degree indicating the maximum value at each measurement point. Line concentration degree calculating means for outputting the degree and / or the direction of the search line at that time,
(3) A line concentration degree image filter comprising line extraction means for determining line information when the line concentration degree calculated by the line concentration degree calculation means is greater than a predetermined noise tolerance value.

Images