JP2004302519A - Object detection method by fractal dimension - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an object detection method by a fractal dimension for detecting a target object accuratly or with an enhanced detecting precision. <P>SOLUTION: Specified areaa are slid (S22) on an image input (S21), and fractal dimension calculation is executed for each specified area (S3). Next, fractal dimensions at each specified area is compared with a fractal dimensional value database (S12), and object areas within a fractal dimension confidence interval are detected (S13). If these processes are executed for the whole of the image input, the detected object areas are integrated (S15). Then, object candidates are ranked in the descending order of the integration number. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an object detection method using a fractal dimension, and more particularly, to an object detection method for recognizing and detecting an object from a natural image using a fractal dimension.
[0002]
[Prior art]
Fractal dimension is a concept that characterizes a continuous, undifferentiated curve. When a graphic is composed of graphic P number which has been reduced in its entirety 1 / N, a suitable index D, expressed as P = N ^D. This exponent part D is a fractal dimension.
[0003]
The fractal dimension is a numerical value representing the characteristics of the fractal, and is also used as a scale representing the complexity of the shape of the object. The larger the fractal dimension, the more complicated the target shape.
[0004]
For example, as shown in FIG. 7A, when a certain figure 10 is divided into 1/6 in length and width, if the figure p1 belongs to six divided blocks, the P = 6 and N = 6 Therefore, the index D = 1. Further, the Koch curve p2 as shown in FIG. 3B has an index D = 1.26186.
[0005]
As described above, conventionally, a method of quantifying and analyzing margin information from a screen using characteristics of a fractal dimension has been developed.
[0006]
[Problems to be solved by the invention]
However, in the above-described conventional method, an object is detected only by a numerical value called a fractal dimension value. Therefore, when detection is performed with a wide confidence interval, a neighborhood or an unrelated area of the target object is also detected. There is a fear that. That is, there is a problem in detection accuracy and detection reliability.
[0007]
The present invention has been made in view of the above-described problems of the related art, and an object of the present invention is to provide a method for detecting an object using a fractal dimension that enables a target object to be detected accurately or with increased detection accuracy. It is in.
[0008]
[Means for Solving the Problems]
In order to achieve the above object, the present invention provides an object detection method using a fractal dimension for detecting an object designated in advance from an image, the object detection method having a predetermined size for a part or whole of an image input. Sliding the designated area to determine a fractal dimension for each designated area; comparing the fractal dimension found for each designated area with the fractal dimension confidence interval of the fractal dimension value database; It is characterized in that it comprises a step of detecting an area as a corresponding object area and a step of integrating overlapping areas of the corresponding object.
[0009]
According to this feature, even if the detection is performed with a wide confidence interval, the specified object can be detected accurately or with increased detection accuracy. Alternatively, an area unrelated to the designated object can be accurately excluded.
[0010]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, the present invention will be described in detail with reference to the drawings. First, a procedure for creating a fractal dimension object database will be described with reference to FIG.
[0011]
In step S1, an image is input. As the image, an image containing a human face, hands, a car, and the like is used. In step S2, a designated area (or an arbitrary side length area) having a predetermined size smaller than the image 1 is slid. This slide is preferably performed on the entire surface of the image 1. In step S3, the fractal dimension of each designated area is calculated.
[0012]
In this step S3, specifically, the input digital image can be divided into an arbitrary side length region, and the fractal dimension can be calculated from the luminance value in the region. Alternatively, an edge filter is applied to the luminance component of the image 1 to calculate a fractal dimension of a feature object region designated as an object after binarization. Note that a Laplacian filter can be used as the edge filter, but another filter may be used.
[0013]
In step S4, a fractal dimension value database is created. At this time, if the object information designation in step S5 is A, B, C, for example, if A = face, B = hand, C = car, the face fractal dimension database D1, the hand fractal dimension database D2, A fractal dimension database D3 of the car is obtained.
[0014]
FIG. 2 is a graph (histogram) of the fractal dimension database D1 of the face image, in which the horizontal axis represents the fractal dimension and the vertical axis represents the frequency. From the figure, it can be seen that when the object is a face, the fractal dimension has a peak around 1.4. As the confidence interval, for example, the fractal dimension can be set to 1.3 to 1.5.
[0015]
Next, a method of detecting a fractal dimension object using the database obtained as described above will be described with reference to FIGS.
[0016]
In step S11, object information to be detected is specified. In step S12, a fractal dimension value database is selected. For example, if the object information is a face, the A object fractal dimension database D1 is selected.
[0017]
On the other hand, in step S21, an arbitrary image for which an object is to be detected, for example, image 1 in FIG. 4 is input. In step S22, as in step S2, an operation of sliding the designated area (or an arbitrary side length area) 2 is performed. In step S23, the fractal dimension of each designated area associated with the slide is calculated. In step S13, the fractal dimension obtained in step S23 is compared with the A-object fractal dimension database D1, and an object area whose fractal dimension belongs to a predetermined confidence interval, for example, 1.3 to 1.5 is detected. You. In step S14, it is determined whether the slide of the designated area has been completed over the entire area of the image input. If this determination is negative, the process proceeds to step S22, in which the specified area is slid. The slide in the designated area may be a part of the image input area instead of the entire area. Hereinafter, the above-described processing is repeated.
[0018]
If the determination in step S14 becomes affirmative as a result of the repetition of this process, the operation proceeds to step S15 to perform an operation of integrating the same object overlapping area. That is, when the determination in step S14 is affirmative, a plurality of object regions detected in step S13 are generally detected. For example, as shown in FIG. 5, a plurality of object areas are detected. When the confidence interval of the feature fractal dimension value is specified widely, a shape object as complicated as a face, for example, a hair, a design of clothes, and wrinkles may be erroneously extracted.
[0019]
Thus, in step S15, the same object overlapping area is integrated. In this integration processing, the higher the number of integrated objects, the higher the priority. That is, as shown in FIG. 3, ranking is performed based on the number of integrations of the detected object areas, and the one with the largest number of integrations is set as the first object candidate, and the one with the second largest number of integrations is the object candidate 2 The object having the next highest number of integrations is the third object candidate. Note that the fractal dimension value information obtained in step S23 is added to each object area to be integrated.
[0020]
When the object overlapping areas are integrated as described above, for example, a face object as shown in FIG. 4B can be detected accurately or with increased detection accuracy. In other words, a region unrelated to the object can be ignored with high precision.
[0021]
Next, when the number of object areas to be integrated is the same and the object candidates cannot be ranked, the priority is set higher as (| average fractal dimension value of integrated object−characteristic fractal dimension value |) is smaller. For example, as shown in FIG. 6, a difference between the central value of the fractal dimension value of the database (1.4 in the above example) and the average value of the fractal dimension value of the integrated area is calculated, and the difference is calculated. The smaller one is ranked higher and the larger one is ranked lower.
[0022]
By the above processing, for example, the possibility of erroneously detecting the screen of the hair, the design of the clothes, and the wrinkles for the target object on the face screen is reduced.
[0023]
【The invention's effect】
As is apparent from the above description, according to the present invention, the overlapping area of the corresponding object detected by comparison with the fractal dimension confidence interval of the fractal dimension value database is integrated, so that it is accurate or with high accuracy. , The target object can be detected. Alternatively, a region unrelated to the target object can be ignored with high accuracy.
[0024]
Further, according to the present invention, the ranking of the object candidates is performed based on the number of integrated regions, and in the case of the same number, the one closer to the central value of the fractal dimension value is ranked higher. Will be able to do it.
[Brief description of the drawings]
FIG. 1 is a flowchart showing a procedure for creating a fractal dimension object database.
FIG. 2 is a diagram illustrating an example of an image input, an arbitrary side length area, and a detected object area.
FIG. 3 is a diagram in which a fractal dimension database D1 of a face is graphed.
FIG. 4 is a flowchart illustrating an operation of detecting a designated object area according to an embodiment of the present invention.
FIG. 5 is a diagram illustrating an example of a corresponding object detected by comparison with a fractal dimension confidence interval in a fractal dimension value database.
FIG. 6 is a flowchart showing a process when the object candidates are at the same rank.
FIG. 7 is an explanatory diagram of a fractal dimension.
[Explanation of symbols]
1 ... image, 2 ... designated area (arbitrary side length area)

Claims (5)

An object detection method based on a fractal dimension for detecting an object designated in advance from an image,
Sliding a designated area of a predetermined size with respect to a part or the whole area of the image input, and obtaining a fractal dimension for each designated area;
Comparing the fractal dimension determined for each of the designated areas with the fractal dimension confidence interval of the fractal dimension value database, and detecting a designated area falling within the confidence interval as a corresponding object area;
Integrating the overlapping areas of the object in question by a fractal dimension. The object detection method based on fractal dimension according to claim 1,
An object detection method using fractal dimension, wherein an image in the designated area is divided into an arbitrary side length area, and a fractal dimension value is calculated from luminance values in the arbitrary side length area. 3. The method for detecting an object according to a fractal dimension according to claim 1 or 2,
An object detection method based on fractal dimension, wherein a fractal dimension value is calculated after using an edge filter for pixel data in the designated area. 4. The object detection method according to claim 1, wherein the object is detected by a fractal dimension.
A method for detecting objects by a fractal dimension, wherein ranking of candidate objects is performed in descending order of the number of integrations of the overlapping areas. The object detection method based on a fractal dimension according to claim 4,
An object detection method based on fractal dimension, wherein, when the number of integrations of the overlapping areas is equal, a part closer to a central value of fractal dimension values of the database is ranked higher.

Images