JPH096961A - Processing device and method for dividing area - Google Patents

Abstract

PURPOSE: To extract an area in real time by determining a processing start point and moving a node on the prolongation of the straight line connecting the node and processing start point. CONSTITUTION: A feature point detecting circuit 11 detects a feature point from an input signal. A feature point distance calculating means 12 calculates the potential prescribed unequivocally by the distance to each feature point other than the feature point. An initial point detecting circuit 13 determines the processing start point corresponding to the potential at the point. In an expansion node initialization step, plural nodes are generated on the basis of the processing start point. Further, a circle expansion calculating circuit 14 moves the nodes on the prolongation of the straight line connecting the nodes and processing start points. An area processing circuit 16 extracts a closed area encircled with the nodes after the movement. Thus, the processing start point is determined corresponding to the potential at the point and on the basis of the processing start point, the nodes are generated and moved on the prolongation of the straight line connecting the nodes and processing start point, so that the closed area encircled with the nodes after the movement is extracted.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an area division processing apparatus and method for dividing a signal into a plurality of areas according to its characteristics. For example, such as an object identification apparatus and a product sorting apparatus, the analysis and recognition of the signal can be performed. TECHNICAL FIELD The present invention relates to an area division processing apparatus and method suitable for use when extracting an object from an input signal in a signal processing apparatus for performing the processing.

The present invention is also characterized by the characteristics of the area in an encoding device, such as a data transmission device or a data recorder, which efficiently compresses the information amount of an original signal for signal transmission or recording / reproduction. It is applied when processing according to.

[0003]

2. Description of the Related Art For example, a signal area division processing method is as follows.
Calculate the feature amount for each point, identify from the similarity of the feature amount whether adjacent points belong to the same region, and integrate,
Split-and-merge method in which each point is sequentially collected to form a region, or a characteristic vector of a signal is detected, and a potential vector field is calculated by differentiating the distance from the characteristic point so that the sum of potential energy is minimized. There is a snake method that calculates a different shape and uses the inside of the shape as the area.

The split-and-merge method uses point-by-point calculations and is relatively easy to implement, but as point integration progresses, local properties are no longer reflected and the obtained region shape is inaccurate. There is a drawback that Although the processing is complicated, the snake method is often used because it can handle the distribution of values as a whole and can adapt it well to the contour of the object.

FIG. 13 shows an example of a signal processing apparatus using the snake method for area shaping. In the signal processing device shown in FIG. 13, the input signal A11 is input to the feature point detection circuit 101, where feature points such as edges are detected, and the feature point signal A12 is output therefrom. The feature point signal A12 is input to the potential calculation circuit 102, where the potential vector of each point of the input signal is calculated, and the potential signal A13 is calculated.
Is output. The calculation method of this potential is shown below.

That is, the two-dimensional potential field P (x, y) when the distance from each edge (feature point) is d (x, y)
Is expressed as follows using the unit vector i, j.

[0007]

[Equation 1]

The initial point detection circuit 103 uses the potential signal A13
The maximum point of the potential is detected from, and the coordinates of the maximum point are marked as the initial point in the descending order of the potential at the maximum point, and the initial point potential signal A14 is output.
This initial point potential signal A14 is the snake calculation circuit 104
, A closed curve having a shape corresponding to the potential distribution is generated by a snake algorithm described later with reference to FIG. 14, and a closed curve shape signal A15 is output. The area processing circuit 105 receives the input signal A11 and the closed curve shape signal A1.
A signal surrounded by a closed curve is extracted from 5, signal processing is performed by an appropriate method, and the result is output as a region processing signal A16. The feature point signal A12 and the region processing signal A16 are multiplexed circuits.
It is multiplexed by 106 and output as an output signal A17.

Next, the calculation method of the Snake algorithm will be shown below by mathematical expressions.

The energy E is defined as follows for the point sequence {e _i } = {(x _i , y _i )}.

[0011]

[Equation 2]

Here, e _ext (i) is the potential energy in the snake's potential field, and e _int (i) is the tension energy due to the shape of the snake. These are vectors {v connecting the potential fields P (x, y) where the distance from the edge (feature point) is d (x, y) and each node of the snake.
_i } is used to express each of the following.

[0013]

(Equation 3)

In the Snake algorithm, {(x _i , y _i )} that minimizes the energy E in the equation (2).
Ask for a pair of.

FIG. 14 is a flow chart showing the snake algorithm in the snake calculation circuit 104 shown in FIG.

The input initial point potential signal in FIG. 14 corresponds to the initial point potential signal A14 in FIG. In the initial point inspection step 111, the initial points marked in the input initial point potential signal A14 are checked in ascending order, and if there is an initial point, the initial point having the largest potential at present is passed to the snake setting step 112. . If the initial point does not exist, it is considered that the configuration of all the snakes has been completed, and a list of all the configured snake node points is output. This snake node list corresponds to the closed curve shape signal A15 in FIG.

In the snake setting step 112, the initial point is set as the center point of snake generation, and in the snake node initialization step 113, the initial nodes are arranged at appropriate intervals from the snake center point. In the potential calculation step 114,
The potential energies at all the nodes indicated by the initial nodes are obtained, and the sum is calculated. In the simultaneous equation calculation step 115, from this sum of potential energies,
In order to minimize the energy E in equation (2), {(x _i , y
_i )} is solved to solve the simultaneous equations and the node movement amount is calculated. In the node movement processing step 116, a new node whose coordinates are moved with respect to each node by the amount of node movement is generated.

In the snake convergence judgment step 117, the new node is compared with the original node, and if the movement amount is larger than a certain threshold value, it is considered that the snake has not converged yet, and the new node is set to the original node. The node is replaced and the process returns to the potential calculation step 114, and the amount of movement of the node is calculated again from the potential. If the amount of movement is less than a certain threshold value, it is considered that the snake has converged, the snake process centered on this initial point is terminated, and the approximate node is output to the in-region point inspection step 118.

Finally, in the in-region point inspecting step 118, the initial points existing in the region defined by the approximate nodes are erased to be excluded from the snake center candidates, and the process returns to the determination in the initial point inspecting step 111 as a definite node.

[0020]

In a region division processing method for detecting a feature point of a signal and extracting a region surrounded by the feature point as a single region, a snake algorithm by minimizing potential energy as described above. When using, the calculation of the potential field and the amount of calculation of the simultaneous equations for finding the energy minimum point are enormous, the calculation cost is too high, and more iterations are required to converge. There is a problem of becoming difficult. Further, the energy E in the equation (1) is not uniquely determined by the distance to the feature point, for example, so that it may fall into a minimum potential solution, and an accurate region shape may not be obtained. There is a problem that can happen.

The present invention has been made in view of such a situation, and makes it possible to perform real-time processing and obtain an accurate region shape.

[0022]

According to a first aspect of the present invention, there is provided an area division processing device which is uniquely defined by a detection means for detecting a feature point from an input signal and a distance to each feature point other than the feature point. Calculation means for calculating the potential to be processed, determination means for determining the processing start point corresponding to the potential of the point, generation means for generating a plurality of nodes based on the processing start point, and processing of the contact point with the contact point. It is characterized by comprising a moving means for moving on an extension of a straight line connecting the starting point and an extracting means for extracting a closed region surrounded by the node after the movement.

According to a seventh aspect of the present invention, the area division processing method detects a feature point from an input signal, calculates a potential uniquely defined by the distance to each feature point other than the feature point, and calculates the potential of the point. Determine the processing start point according to the potential, generate multiple nodes based on the processing start point, move the node on the extension of the straight line connecting the contact and the processing start point, and The method is characterized by extracting a closed region that is surrounded.

The area division processing apparatus according to the eighth aspect of the present invention provides a selection means for selecting an intermediate point which is a point on the closed curve and different from the node, and a value determined from the positions of the intermediate point and the feature point. And threshold value processing, and processing means for adding an intermediate point as a node in accordance with the result.

The area division processing method according to claim 11 is
Select a midpoint as a point on the closed curve that is different from the node, threshold the value determined from the position of the midpoint and the feature point, and set the midpoint as the node according to the result. It is characterized by adding.

The area division processing device according to claim 12 is
The calculation means for calculating the center point of the area surrounded by the closed curve formed by the nodes after the movement processing, the detection means for detecting the change point from the change in the distance from the center point of the node, and the detection means Interpolation means for interpolating the change point in accordance with the detection result is provided.

The area division processing method according to claim 18 is
The center point of the area enclosed by the closed curve formed by the nodes after the movement processing is calculated, the change point is detected from the change in the distance from the center point of the node, and the change is detected according to the detection result. It is characterized by interpolating points.

The area division processing device according to claim 19 is
From the input signal, a non-closed curve region which is a region formed by points other than the feature points and which is not surrounded by a closed curve is detected, and the processing condition for determining the non-closed curve region as the region surrounded by the closed curve is determined. It is characterized by comprising a judging means and a processing means for executing processing for setting the non-closed curve area to an area surrounded by a closed curve, corresponding to the result of the judgment of the judging means.

The area division processing method according to claim 23 is
From the input signal, a region constituted by points other than the feature points, a non-closed curve region not enclosed by a closed curve is detected, and a condition for processing the non-closed curve region as a region enclosed by a closed curve is determined, It is characterized in that the non-closed curve region is processed as a region surrounded by a closed curve in accordance with the result of the determination.

[0030]

In the area division processing device according to claim 1,
The detecting means detects a characteristic point from the input signal, the calculating means calculates a potential uniquely defined by the distance to each characteristic point other than the characteristic point, and the determining means corresponds to the potential of the point. Then, the processing start point is determined, the generating means generates a plurality of nodes based on the processing start point, and the moving means moves the node on an extension line of a straight line connecting the contact point and the processing start point, Extraction means extracts a closed region surrounded by the moved nodes.

In the area division processing method according to the seventh aspect, the characteristic points are detected from the input signal, and the potential uniquely defined by the distance to each characteristic point other than the characteristic points is calculated, Determine the processing start point according to the potential of the point, generate multiple nodes based on the processing start point, move the nodes on the extension line of the straight line connecting the contact point and the processing start point, and The closed region surrounded by the nodes is extracted.

In the area dividing processing device according to the eighth aspect, the selecting means selects an intermediate point which is a point on the closed curve and is different from the node, and the processing means is characterized by the intermediate point. The value determined from the position of the point is thresholded, and the intermediate point is added as a node corresponding to the result.

In the area division processing method according to the eleventh aspect, an intermediate point which is a point on the closed curve and different from the node is selected, and a value determined from the positions of the intermediate point and the characteristic point is set. Threshold processing is performed, and intermediate points are added as nodes according to the result.

In the area division processing apparatus according to the twelfth aspect, the calculating means calculates the center point of the area surrounded by the closed curve formed by the nodes after the movement processing, and the detecting means causes the nodal points. The change point is detected from the change in the distance from the center point, and the interpolating means interpolates the change point according to the detection result of the detecting means.

In the area division processing method according to the eighteenth aspect, the center point of the area surrounded by the closed curve formed by the nodes after the movement processing is calculated, and the change in the distance from the center point of the node is calculated. Then, the change point is detected and corresponding to the detection result,
Interpolate change points.

According to another aspect of the present invention, in the area division processing device, the judging means detects, from the input signal, a non-closed curve area which is an area constituted by points other than the characteristic points and which is not surrounded by a closed curve. Then, the processing means judges the condition of the processing for making the non-closed curve area into the area surrounded by the closed curve, and the processing means makes the non-closed curve area into the area surrounded by the closed curve in accordance with the result of the judgment of the judgment means. To execute.

In the area division processing method according to the twenty-third aspect, a non-closed curve area which is an area constituted by points other than the feature points and which is not enclosed by a closed curve is detected from the input signal, and the non-closed curve is detected. A condition for processing the area as an area surrounded by a closed curve is determined, and the non-closed curve area is processed as an area surrounded by the closed curve according to the result of the determination.

[0038]

EXAMPLES Examples of the present invention will be explained below, but in order to clarify the correspondence between each means of the invention described in the claims and the following examples The characteristics of the present invention will be described below by adding the corresponding embodiment (however, one example) to the above. However, of course, this description does not mean that each means is limited to the described one.

The area division processing device according to the first aspect is a detection means for detecting a feature point from an input signal (for example, the feature point detection circuit 11 in FIG. 1) and a distance to each feature point other than the feature point. A calculation means (for example, the feature point distance calculation circuit 12 of FIG. 1) for uniquely defining the potential and a determination means (for example, the initial point detection of FIG. 1) for determining the processing start point corresponding to the potential of the point. A circuit 13), a generating means for generating a plurality of nodes based on the processing start point (for example, expansion node initialization step 23 in FIG. 3), and the nodes on an extension line of a straight line connecting the contact point and the processing start point. It is characterized by comprising moving means (for example, the circular expansion calculation circuit 14 in FIG. 1) for moving and extraction means (for example, the area processing circuit 16 in FIG. 1) for extracting the closed area surrounded by the moved nodes. .

The area division processing device according to claim 8 is a selecting means for selecting an intermediate point which is a point on the closed curve and is different from the node (for example, the midpoint detecting step 4 in FIG. 6).
2) and threshold value processing of a value determined from the positions of the intermediate point and the characteristic point, and corresponding to the result, a processing means for adding the intermediate point as a node (for example, a node addition processing step 46 in FIG. 6)
And characterized in that:

The area division processing device according to claim 12 is
From the change of the distance between the center point of the area enclosed by the closed curve formed by the nodes after the movement processing and the area surrounded by the closed curve (for example, distance calculation step 52 in FIG. 10), and the change point of the change point, Detecting means (for example, the protruding point inspection step 54 and the return point inspection step 56 in FIG. 10) for detecting the error, and the interpolation means for interpolating the change point corresponding to the detection result of the detecting means (for example, the correction step 58 in FIG. 10). And is provided.

An area division processing apparatus according to claim 19 is
Judging means (for example, as shown in FIG. 1) for judging processing conditions for defining a non-closed curve area, which is an area constituted by points other than the feature points and which is not surrounded by a closed curve, from the input signal as an area surrounded by the closed curve. Reregion selection circuit 15) and processing means for executing processing for setting the non-closed curve area as an area surrounded by a closed curve in accordance with the result of the judgment by the judgment means (for example, feature point distance calculation circuit 12 in FIG. 1). And is provided.

Examples of the present invention will be described below. FIG. 1 shows a configuration example of a signal processing device according to the present invention.

In the signal processing apparatus shown in FIG. 1, the input signal B11 is input to the feature point detection circuit 11, where feature points such as edges are detected and a feature point signal B12 is output.
The feature point distance calculation circuit 12 calculates the distance from one of the feature point signal B12 and the re-regionized feature point signal B16 described later to the closest feature point from each point as the feature point distance, Output point distance signal B13. This feature point distance represents the distance from each point to the lowest feature point in Euclidean distance. Generally, here, the potential at each point is calculated, but since the potential value in the present invention is equivalent to the distance from each point to the closest feature point, this potential will be appropriately referred to as the distance hereinafter. .

The initial point detection circuit 13 detects the maximum point of the signal value from the characteristic point distance signal B13, marks the coordinates of the maximum points as the initial points in the descending order of the characteristic point distance at the maximum point, and then outputs the initial point distance signal B14. Is output. Initial point distance signal B14
Is input to the circle expansion calculation circuit 14, where a closed curve having a shape closest to the feature point is generated by the circle expansion algorithm, and a closed curve shape signal B15 is output. This circle expansion algorithm will be described later with reference to FIG.

The re-regionalization selection circuit 15 uses the closed curve shape signal B1.
Check if 5 fills the screen (regionalized with a closed curve) and if not, replace the already regionalized points with a mask to reregionalize the rest The re-regionized feature point signal B16 is generated and input to the feature point distance calculation circuit 12. If the screen is filled with the closed curve shape signal B15, the re-regionalization selection circuit 15 terminates the regionization process, and generates and outputs the regionization signal B17.

The region processing circuit 16 extracts a signal surrounded by a closed curve from the input signal B11 and the region-ized signal B17, performs signal processing by an appropriate method, and outputs the result as a region processed signal B18. The feature point signal B12 and the region processing signal B18 are multiplexed by the multiplexing circuit 17 and output as the output signal B19.

Next, the operation will be described. The feature point detection circuit 11 detects a feature point from the input signal B11 and outputs a feature point signal B12 to the feature point distance calculation circuit 12. The characteristic points are points representing edges, for example, as indicated by C11 in FIG. The feature point distance calculation circuit 12 uses this feature point signal B12 to calculate the distance (feature point distance) from each point to the closest feature point. As a result, as shown in FIG. 2, a map constituted by the distance (potential) to the characteristic point C11 is constructed, and the characteristic point distance calculation circuit 12 causes the characteristic point distance representing the potential map by this distance. Signal B13 is the initial point detection circuit 1
Output to 3.

The initial point detection circuit 13 has a feature point distance signal B13.
Detect the maximum point of. As shown in FIG. 2, this maximum point is represented as a maximum point C13 (C13-1 to C13-5) surrounded by an equidistant line C12 representing a position equidistant from the characteristic point C11.

Then, the initial point detection circuit 13 determines the maximum point
This is marked as an initial point in descending order of the feature point distance in C13 (in the embodiment of FIG. 2, C13-1, C13-2, C13-
3, C13-4, C13-5 in order), initial point distance signal B
It is output to the circle expansion calculation circuit 14 as 14. The circle expansion calculation circuit 14 processes the initial point distance signal B14 according to a circle expansion algorithm.

Next, referring to the flowchart of FIG.
The circle expansion algorithm will be described.

Generally, as explained in the conventional example,
It is necessary to calculate the potential at each point,
In this embodiment, the potential value is equivalent to the distance from each point to the closest feature point. The feature point distance is the Euclidean distance representing the distance to the closest feature point at each point. Of course, it is also possible to adopt the potential used in the description of the conventional example.

The initial point distance signal as an input in FIG. 3 corresponds to the initial point distance signal B14 in FIG. In the initial point inspection step 21, the initial points marked in the input initial point distance signal B14 are checked in ascending order, and if there is an initial point, the initial point having the largest distance (see FIG.
In the case of, C13-1) is passed to the initial circle setting step 22.

In the initial circle setting step 22, the initial point (maximum point having the largest feature point distance) is set as the center point of the initial circle of the circle expansion process, and the initial circle whose radius r is the feature point distance at the center of the initial circle is set. Set. For example, as shown in FIG. 4, a central maximum point is set as an initial circle center point C13-1, and an initial circle C14 is set at a position having a radius r from that point.

In the expansion node initialization step 23, the initial nodes are arranged at appropriate intervals from the initial circle center point. In the embodiment of FIG. 4, the initial circle C14 having the radius r is divided by the angle of θ = π / 2r to be the initial nodal point interval C15.

In the distance calculation step 24, the node distance of the initial node C16 (distance from the initial circle center point (in this case,
r)). In the movement inspection step 25, when the coordinates of each node are expanded in the radial direction of the initial circle C14 by the node distance (r) at that point (the initial circle center point C13 and the initial node C16
On the extension of the straight line connecting to and, when it is moved away from the center point C13), it is checked whether or not it hits the feature point C11 (whether it touches). In the case of collision, it is assumed that this node is not moved any more, and the process returns to the distance calculation step 24, and the process is continued for the next node. If there is no collision, the node movement amount (r) is output. In the nodal point movement processing step 26, a new nodal point whose coordinates are moved with respect to the nodal point by the nodal point movement amount (r) is generated.

In the approximation adjustment processing step 27, the approximation adjustment is performed on the new node so that the new node can be fitted well to the arrangement of the feature points around the new node, and the new approximate node is generated. This approximation adjustment algorithm will be described later with reference to FIG.

In the convergence judgment step 28, this approximate new node is compared with the original node, and if the movement amount is equal to or greater than a threshold value, it is considered that the node has not converged yet, and the original node is replaced with the approximate new node. Then, the process returns to the distance calculation step 24, and the movement amount of the approximate new node is calculated again from the node distance. If the amount of movement is less than a certain threshold value, it is considered that the movement has converged, and the circular expansion processing centered on this initial point is terminated.

As described above, for example, as shown in FIG. 4, the distance r from the initial circle center point C13-1 to a predetermined initial node C16 on the initial circle C14 is set as the node distance, and the initial node C16 is set.
Is the initial circle center point C13 and its initial node
It is determined whether or not it is possible to move on an extension of the straight line connecting with C16. Even if the initial node C16 is moved by the node distance r, if the new position does not contact the feature point C11, the initial node C16 moves (expands) to the new position.
Is done. Such an operation is repeatedly executed until the movement amount becomes equal to or less than the predetermined threshold value. Thus, as shown in FIG. 4, the original initial node C16 is moved by the node moving amount C17 to become the new node C18.

The above processing is performed by the initial circle center point C13-1.
For each initial node C16 on the initial circle C14 centered at. As a result, as shown in FIG. 5, the initial circle C14 is expanded, and the closed node curve C19 is generated by the new node C18.

In this way, the convergence judgment step 28 outputs the approximate node to the consistency check step 29. In the consistency check step 29, the abnormality of the approximate node due to the mismatch of the feature points is corrected, and the corrected approximate node is output. This consistency check algorithm will be described later with reference to FIG.

Finally, in the in-region point inspection step 30, the initial center point C13-1 existing in the closed region curve C19 formed by the modified approximate node C18 is erased and removed from the candidates as the initial circle center point. , The new node C18 on the closed region curve C19 is the fixed node. Then, the process returns to the determination of the initial point inspection step 21.

In the initial point inspection step 21, the next largest maximum point is selected as the initial circle center point, and the processing after the initial circle setting step 22 is performed in the same manner as described above. Then, in the initial point inspection step 21, if it is determined that there is no other initial point, the signal related to the closed region curve C19 obtained by the processing up to that point is output as the closed curve shape signal B15 (on the closed region curve C19. The new node list of C18 is output as an expanded node list).

Next, the approximate adjustment algorithm of the above-mentioned approximate adjustment step 27 of FIG. 3 will be described with reference to the flowchart of FIG.

The input signal in FIG. 6 is the above-mentioned new node C18.
Is equivalent to In the nodal loop control step 41, it is judged whether or not the inputted nodal point is the last nodal point of the closed region curve C19, and if it is not the last nodal point, the process proceeds to the next midpoint detecting step 42.

That is, by the processing up to that point, FIG.
As shown in (A), the initial node C16 set on the initial circle C14 centered on the initial circle center point C13 is replaced by the initial circle center point C1.
Move to the line connecting 3 and its initial node C16,
A closed region curve (approximate curve) C19 composed of a new node (node after movement) C18 is generated as shown in FIG. If the new node C18 on the closed region curve C19 is not the last node, the process proceeds to the midpoint detection step 42.

In the midpoint detection step 42, as shown in FIG. 7C, the coordinates of the midpoint M between the predetermined node A _i and the adjacent node A _{i + 1} are calculated, and the inter-feature points are calculated. In the distance calculation step 43, the coordinates of the point N at which the straight line extending from the initial circle center point C13 through the middle point M hits the characteristic point C11 are calculated, and the distance LMN between the middle point M and the point N is calculated. Further, the distance calculation step 44
Then, the node distance of the midpoint M (distance from the initial circle center point C13 to the midpoint M) dM is obtained. In the proximity check step 45, the distance
When the distance LMN is larger than the distance dM by comparing LMN with the distance dM, the point N is added as a new node by the node addition processing step 46 in order to obtain an accurate approximation.

When the distance LMN is equal to or smaller than the distance dM, the processing of the node addition processing step 46 is skipped.

As described above, when the processing for all the new nodes C18 of the closed region curve C19 is completed, the same processing is repeatedly executed until it is determined in the node loop control step 41.

In the embodiment of FIG. 7, the point A _i and the midpoint M of the point A _{i + 1} adjacent to the point A _i are obtained as shown in FIG. The point at which the straight line C21 connecting the initial circle center point C13 and the middle point M contacts the feature point C11 is N. The distance dM from the initial circle center point C13 to the middle point M and the distance LM between the points M and N
When N is compared and the distance LMN is larger than the distance dM, as shown in FIG. 7D, the point N is the new node C1 of the closed region curve C19.
Added as 8. In other words, the closed region curve C19 is shown in FIG.
The state shown in (B) is changed to the state shown in FIG.

In this way, a new node is provided between the nodes that could not be sufficiently approximated to the characteristic point C11 by simply expanding each node, and the approximation is adjusted. As a result, a closed region curve (approximation curve) C19 approximated by the feature point C11 can be obtained. A signal including the nodes added to match the approximate shift between the nodes is output as the approximate adjustment node signal.

It is to be noted that the distance (potential) to the characteristic point C11 at each point between two adjacent node points C16, which is the point on the closed region curve C19, is calculated, and the point having the maximum distance is set to the center. The point M may be selected instead of the point M and the same processing may be performed.

By the way, depending on the arrangement of the detected feature points, each node approximated by the above-mentioned processing converges to a feature point for a region different from the feature point enclosing the original region, and is obtained. The shape of the area may be greatly distorted. For example, as shown in FIG. 8, there is a break in the edge (feature point C11) surrounding the area, and the approximate node C3
3, C34 may converge to another outer edge (feature point C11A).

Nodes C33 and C3 that have protruded in this way
In order to arrange 4 inside the edge that would originally exist, as shown in FIG. 8, the protruding nodes C33, C34 and other nodes between them are deleted, and the nodes before and after that are deleted. It is possible to reconnect the line segment (straight line) C31 between the nodes C32 and C35. Alternatively, as shown in FIG. 9, a curve C41 is drawn between the nodes C32 and C35 that did not extend, and the curve C41 is pulled back to a position where a straight line connecting the center point C13 and the node C33 intersects, Similarly, at the position where the curve C41 and the straight line connecting the center point C13 and the node C34 intersect, the node C34
Can be pulled back.

The processing in the consistency checking step 29 shown in FIG. 3 is the one shown in FIG. 8 or 9. The flowchart of FIG. 10 shows a processing example in this case.

In the node loop control step 51, it is judged whether or not it is the last node on the predetermined closed region curve C19, and if it is not the last node, the process proceeds to the distance calculation step 52 of the next stage. In the distance calculation step 52, the center point C13 is determined,
Calculate the distance R _i between each node A _i and the center point C13. In the difference calculation step 53, a node having the smallest distance from the center point C13 (a node represented as A _{i in} FIGS. 8 and 9) is set, and the distance R _i of each node A _{i is} sequentially calculated from the starting point. The difference d _i of the distance R _{i-1 from} the previous node A _i-1 is calculated.

In the protruding point inspection step 54, the difference d _i is checked, and if the difference d _i is larger than the threshold value Th and the protruding point flag is not set, the protruding point flag is set in the protruding point setting step 55. Is set and this node number is stored as the protruding point. In addition return point test step 56, examines the difference d _i, the difference d _i is less than the negative threshold Th, and if protruding point flag is set, the return point set step 57, the return point flag Is set and this node number is stored as a return point.

The above processing is performed between all the nodes on all closed region curves C19 to detect all the pairs of the protrusion point and the return point, and then in the correction step 58, the protrusion point flag and the return point flag are referred to. Then, the node is corrected between each pair of the protrusion point and the return point by the interpolation processing as shown in FIG. 8 or 9.

As described above, for example, as shown in FIG. 11 (A), when the distance R _{i of} each node A _{i from} the center point C13 is obtained, as shown in FIG. 11 (B), the node A _i distance R _i
And the difference d _i between the previous node A _{i-1 and} the distance R _i-1 is obtained.
Then, this difference d _i is compared with the threshold Th.

When the difference d _i is larger than the threshold value Th, the node A _i is stored as the protruding point like the node C33 in FIGS. 8 and 9 and stored.

Similarly, when it is determined that the difference d _i is smaller than the negative threshold value −Th, the node number is stored as a return point like the node C34 shown in FIGS. 8 and 9.

When the pair of the protrusion point and the return point is stored in this way, the protrusion point and the return point are deleted, and the node and the return point immediately before the protrusion point are deleted as shown in FIG. 11C, for example. The process of complementing with the node immediately after is executed. That is, as shown in FIG. 8, the node C32 immediately before the node C33 as the protrusion point and the node C32 as the return point
Connect the node C35 immediately after C34 with the line segment C31.
33 and C34 are deleted.

[0083] Alternatively, as shown in FIG. 12 (A), when the center point C13 distance R _i of each node A _i is obtained, the distance of the node A _i R _i and, immediately preceding node A _i- difference d _i between the _first distance R _i-1 is obtained as shown in FIG. 12 (B). This difference d _i is the positive threshold Th and the negative threshold -T.
By comparison with h, as shown in FIG. 9, a node C33 as a jump-out point and a node C34 as a return point are obtained.

Then, as shown in FIG. 12C, a curve is drawn between the node immediately before the protruding point and the node immediately after the return point, and the nodes on the curve are interpolated.

That is, as shown in FIG. 9, a curve C41 is drawn between the node C32 immediately before the node C33 as the protruding point and the node C35 immediately after the return point C34, and the curve C41, the center point C13 and the node The node C33 is moved to the intersection with the straight line connecting the C33, and the node C34 is moved to the intersection of the curve C41 and the straight line connecting the center point C13 and the node C34.

As the curve C41, a straight line can be used in addition to a circular arc, a spline curve and the like.

In this way, it is possible to prevent the nodes from converging on the characteristic points of the area different from the characteristic points that surround the original area, and the shape of the obtained area being greatly distorted. .

In the distance calculation step 52 of FIG. 10, the center point of the closed region curve C19 formed by the moved nodes may be calculated. At this time, as in the case of the process shown in the flowchart of circular expansion in FIG. 3, the maximum points of the distance to the feature point are ordered in the order of the magnitude of the distance, and the maximum point is selected as the center point in that order. ,
It is possible not to select the maximum point included in the determined area as the center point.

As described above, the circular expansion calculation circuit of FIG.
The process in 14 is performed, and the closed area curve (approximate curve) C19
A closed curve shape signal B15 corresponding to is output. The re-region selection circuit 15 determines whether or not all the signals on the screen as shown in FIG. 2, for example, are filled with the closed curve shape signal B15, and a region (closed curve shape signal B15 is not yet generated ( If there is a non-closed curve region), the re-regionalized feature point signal B16 that masks the already regionalized points
Is generated and is output to the feature point distance calculation circuit 12. Each of the circuits after the feature point distance calculation circuit 12 executes the region formation processing in the same manner as the above case.

As a result, for example, the maximum point C13- shown in FIG.
The process of approximating the region surrounded by the feature points C11 including 1 to C13-5 with the closed region curve C19 is sequentially executed.

When the region forming process for all the regions surrounded by the feature point C11 is completed, the reregion forming selecting circuit 15
The closed curve shape signal B15 supplied from the circular expansion calculation circuit 14 is output to the region processing circuit 16 as a regionization signal B17.

The area processing circuit 16 extracts the signal of the area surrounded by the closed curve from the input signal B11 and the area-ized signal B17,
Perform signal processing by an appropriate method. For example, the data is compressed for each area, and if it is image data, the image is recognized. The processing result is output to the multiplexing circuit 17 as a region processing signal B18. The multiplexing circuit 17 multiplexes the area processed signal B18 input from the area processing circuit 16 with the feature point signal B12 input from the feature point detection circuit 11,
Output as output signal B19.

In the above embodiment, a circle is set as the initial shape, but other than this, for example, an arbitrary closed polygon or closed curve may be set. Also, different initial shapes may be set according to the potential and the distance.

Further, in the above embodiment, the Euclidean distance is used as the potential, but in addition to this, for example, the chessboard distance (the number of the longer eyes is the distance) and the city area distance (vertical direction and horizontal direction). An arbitrary distance calculation method may be used, such as the sum of the sides of the eyes being the distance), and the potential obtained by taking the first derivative of these may be used. In short, any potential that is uniquely defined by the distance (potential that monotonically increases or monotonically decreases) may be used.

In the above embodiment, the determination of the re-regionalization process is made based on the presence / absence of the closed curve region. However, in addition to this, for example, the region of the non-closed curve region is never performed, is always performed a predetermined number of times. If it exceeds a certain threshold, do it,
You may make it judge by the criteria (conditions) like this.

As described above, by limiting the movement of the nodes to the movement along the straight line and by performing the sequential correction of the nodes,
The following effects can be obtained. 1. Compared with the conventional convergence algorithm for minimizing potential energy, since the potential is uniquely given by the distance, it does not fall into a minimal solution, it can be processed with a small amount of calculation, and the number of iterations until convergence is reduced. Can be made. 2. Compared with the conventional Snake algorithm, new nodes can be added at locations deviating from the potential, so the area shape can be approximated with high accuracy.

[0097]

As described above, according to the area dividing processing apparatus and the area dividing processing method of the present invention, it is possible to deal with the potential uniquely defined by the distance to the feature point. Since the processing start point is determined and the node is moved on the extension line of the straight line connecting the node and the processing start point, it is possible to extract the area in real time. In addition, since it does not fall into a minimal solution, it is possible to approximate the shape of an accurate region.

According to the area division processing device and the area division processing method of the eleventh aspect, an intermediate point which is a point on the closed curve and different from the node is selected, and the intermediate point is selected. Since the points are added as nodes, the area shape can be approximated more accurately.

According to the area division processing apparatus and the area division processing method of the eighteenth aspect, the center point and the nodes of the area surrounded by the closed curve formed by the nodes after the movement processing are formed. Since the change point is detected from the change in the distance and the change point is interpolated according to the detection result,
It is possible to prevent the shapes from converging on different feature points.

According to the area division processing apparatus and the area division processing method of the twenty-third aspect, the condition for processing the non-closed curve area as the area surrounded by the closed curve is judged, and the result of the judgment is made. Accordingly, the non-closed curve area is processed as an area surrounded by the closed curve, so that the area approximated to the feature point can be determined quickly and reliably.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration example of a signal processing device to which an area division processing device of the present invention is applied.

FIG. 2 is a diagram showing a configuration example of a potential map based on distance.

FIG. 3 is a flowchart showing a process of the embodiment of FIG.

FIG. 4 is a diagram illustrating an initial state of circular expansion processing according to the present invention.

FIG. 5 is a diagram illustrating a convergent state of the circular expansion processing according to the present invention.

FIG. 6 is a flowchart showing a process of an approximation adjustment process step 27 of FIG.

FIG. 7 is a diagram illustrating the algorithm of FIG.

FIG. 8 is a diagram showing a state of protrusion and a matching process due to mismatch of circular expansion according to the present invention.

FIG. 9 is a diagram showing a state of protrusion due to mismatch of circular expansion and another matching process according to the present invention.

10 is a flowchart showing a process of a consistency checking step 29 in FIG.

11 is a graph illustrating the matching process of FIG.

12 is a graph illustrating the matching process of FIG.

FIG. 13 is a block diagram showing a configuration example of a conventional area shaping signal processing device using a snake system.

FIG. 14 is a flowchart showing a snake system process.

[Explanation of symbols]

 11 Feature Point Detection Circuit 12 Feature Point Distance Calculation Circuit 13 Initial Point Detection Circuit 14 Circle Expansion Calculation Circuit 15 Reregion Selection Circuit 16 Region Processing Circuit 17 Multiplexing Circuit

Claims (23)

[Claims] 1. A detection unit for detecting a feature point from an input signal, a calculation unit for calculating a potential uniquely defined by a distance of each point other than the feature point to the feature point; Determining means for determining a processing start point corresponding to the potential; generating means for generating a plurality of nodes based on the processing start point; extension of a straight line connecting the contact point with the processing start point; An area division processing device comprising: a moving means for moving along a line; and an extracting means for extracting a closed area surrounded by the node after the movement. 2. The area division processing apparatus according to claim 1, wherein the calculation means calculates the potential corresponding to a distance to the closest feature point of points other than the feature point. . 3. The determining means searches for maximum points in the distribution of the potential, and orders the points as the maximum points in descending order of the value of the distance to the feature point, and corresponds to the ordering. 2. The area division processing apparatus according to claim 1, wherein the local maximum point is selected as the processing start point and the local maximum point included in the confirmed area is not selected as the processing start point. 4. The generating means divides the circumference of a circle whose center is the processing start point and whose radius is a value determined from the potential of the center point into a plurality of values defined by the radius. The area division processing device according to claim 1, wherein a point is the node. 5. The moving means moves the node generated by the generating means in a radial direction toward the outside of the circle by a length defined by a potential value of the node. The area division processing device according to claim 1. 6. The extracting means generates a closed curve by connecting curved lines between the nodes that are generated and moved with the same processing start point as a reference, and the position and shape of the closed curve are obtained from an input signal. The area division processing apparatus according to claim 1, wherein the signal is extracted according to the following. 7. A feature point is detected from an input signal, a potential uniquely defined by a distance of each point other than the feature point to the feature point is calculated, and processing is performed corresponding to the potential of the point. A start point is determined, a plurality of nodes are generated based on the processing start point, the node is moved on an extension of a straight line connecting the contact point and the processing start point, and the node is surrounded by the moved node. A region division processing method, characterized in that a closed region to be extracted is extracted. 8. A feature point is detected from an input signal, a plurality of nodes are generated, a closed curve connecting the nodes is set, the node is moved so as to approach the feature point, and the node is surrounded by the closed curve. In a region division processing device for determining a region, a selection means for selecting a midpoint as a point on the closed curve that is different from the node, and a value determined from the positions of the midpoint and the feature point A region dividing processing device, comprising: a threshold value process; and a processing unit that adds the intermediate point as the node according to the result. 9. The area division processing apparatus according to claim 8, wherein the selection means selects a midpoint between adjacent nodes on the closed curve as the midpoint. 10. The selecting means calculates a potential at each point between the adjacent nodes on the closed curve,
The area division processing apparatus according to claim 8, wherein a point having the maximum potential is selected as the intermediate point. 11. A feature point is detected from an input signal, a plurality of nodes are generated, a closed curve connecting the nodes is set, the node is moved so as to approach the feature point, and the node is surrounded by the closed curve. In a region division processing method for determining a region, a midpoint as a point on the closed curve that is different from the node is selected, and a value determined from the positions of the midpoint and the feature point is thresholded. Then, in accordance with the result, the intermediate point is added as the node, and the area division processing method is characterized. 12. A feature point is detected from an input signal, a plurality of nodes are generated, a closed curve connecting the nodes is set, the node is moved so as to approach the feature point, and the node is surrounded by the closed curve. In a region division processing device for determining a region, a calculation means for calculating a center point of a region surrounded by the closed curve formed by the nodes after being subjected to movement processing, and a change in distance between the nodes and the center point From the above, there is provided a region division processing device comprising: a detection unit that detects a change point; and an interpolation unit that interpolates the change point in accordance with a detection result of the detection unit. 13. The detecting means includes a distance from the center point of the first nodal point and a second distance adjacent to the first nodal point.
The difference between the distance from the center point of the node of the, the threshold processing in the increasing direction and the decreasing direction is performed, and the node in which the distance increases above the threshold value 13. The area division processing device according to claim 12, wherein the nodal point is detected as an increasing point, and the node at which the distance decreases by the threshold value or more is detected as a decreasing point among the changing points. 14. The area division processing apparatus according to claim 13, wherein the interpolation means deletes the increasing point, the decreasing point, and the nodes between them. 15. The position where the interpolating means intersects a curve connecting the node immediately before the increasing point and the node immediately after the decreasing point and a straight line connecting the increasing point and the center point. A position where the curve connecting the node immediately before the increase point and the node immediately after the decrease point and the straight line connecting the decrease point and the center point intersect each other while moving the increase point. 14. The area division processing apparatus according to claim 13, wherein the decrease point is moved to the second position. 16. The calculating means searches for a maximum point of a potential distribution uniquely defined by a distance to each of the feature points other than the feature point, and selects one of the points as the maximum point. , Ordering in descending order of the value of the distance to the feature point, select the maximum point as the center point corresponding to the ordering, and do not select the maximum point included in the confirmed region as the center point. 13. The area division processing device according to claim 12, wherein: 17. The area division processing apparatus according to claim 13, wherein the detection unit sequentially performs threshold value processing from the starting point with the node having the smallest distance from the center point as a starting point. . 18. A feature point is detected from an input signal, a plurality of nodes are generated, a closed curve connecting the nodes is set, the node is moved so as to approach the feature point, and the node is surrounded by the closed curve. In a region division processing method for determining a region, a center point of a region surrounded by the closed curve formed by the nodes after being moved is calculated, and a change in a distance between the node and the center point is changed. A region division processing method, which comprises detecting points and interpolating the change points in accordance with the detection result. 19. A feature point is detected from an input signal, a plurality of nodes are generated, a closed curve connecting the nodes is set, the node is moved so as to approach the feature point, and the node is surrounded by the closed curve. In a region division processing device for determining a region, from the input signal, a region constituted by points other than the characteristic points, detects a non-closed curve region not enclosed by the closed curve, the non-closed curve region, Determination means for determining a condition of processing to be an area surrounded by the closed curve, and processing for performing processing of making the non-closed curve area an area surrounded by the closed curve, corresponding to the result of the determination by the determination means. An area division processing device comprising: 20. The determination unit according to claim 19, wherein the determination unit determines whether or not the number of times of processing for defining the area surrounded by the closed curve has reached a predetermined number set in advance. Area division processing device. 21. The determining means calculates an area of a region surrounded by the closed curve, thresholds the area, and
20. The area division processing device according to claim 19, wherein it is determined whether or not to perform processing for setting the non-closed curve area to be an area surrounded by the closed curve. 22. The area division processing apparatus according to claim 19, wherein the processing means receives an input of the signal of the non-closed curve area in response to the determination result of the determination means. 23. A feature point is detected from an input signal, a plurality of nodes are generated, a closed curve connecting the nodes is set, the node is moved so as to approach the feature point, and the node is surrounded by the closed curve. In a region division processing method for determining a region, from the input signal, a region constituted by points other than the feature points, detects a non-closed curve region not enclosed by the closed curve, the non-closed curve region, A region division processing method, comprising: determining a condition to be processed as an area surrounded by the closed curve, and processing the non-closed curve area as an area surrounded by the closed curve in accordance with a result of the judgment.