JP2010257324A - Image processing apparatus and method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To achieve highly precise object detection suitable for applying image quality improvement processing. <P>SOLUTION: An image processing apparatus for detecting a prescribed object pattern from an image is configured to detect a prescribed partial region to be verified with the object pattern from the image, and to determine whether the extracted partial region is the object pattern, and to evaluate the image characteristics of the partial region determined as the object pattern by this discrimination, and to determine whether the partial region is the object pattern based on the evaluation results. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to an image processing apparatus, an image processing method, and a program, and more particularly, to a technique suitable for use in detecting a specific subject from an image.

  An image processing method for automatically detecting a specific subject pattern from an image is very useful, and can be used for, for example, determination of a human face. Such methods can be used in many areas such as teleconferencing, man-machine interface, security, monitor systems for tracking human faces, image compression, and the like. Such face detection techniques make use of some prominent features (eyes, mouth, etc.) and the inherent geometric positional relationship between those features, or the symmetric features of human faces, human Using facial color features, template matching, neural networks, etc.

  For example, there is a method of detecting a face pattern in an image by a neural network. Image data targeted for face detection is read into the memory, and a predetermined area to be matched with the face is cut out from the read image. Then, the distribution of pixel values in the cut-out area is input, and one output result is obtained by calculation using a neural network.

  At this time, the weight and threshold value of the neural network are learned in advance using face image patterns and non-face image patterns with a huge amount. Judged to be a face. Then, the face is detected from the image by scanning the cutout position of the image pattern to be collated with the face input by the neural network, for example, from the entire image as shown in FIG. Further, in order to cope with the detection of faces of various sizes, as shown in FIG. 3, the read images are sequentially reduced at a predetermined rate, and the above-described face detection scanning is performed on the images.

  In addition, as an example focusing on speeding up the processing, AdaBoost is used to effectively combine many weak classifiers to improve the accuracy of face discrimination, and each weak classifier is configured with a Haar-type rectangular feature. In addition, the calculation of the rectangular feature amount is performed at high speed using the integral image. Also, the discriminators obtained by AdaBoost learning are connected in series to form a cascade type face detector. This cascade-type face detector first uses a simple discriminator at the front stage to remove pattern candidates that are clearly not faces on the spot, and only the other candidates at the back stage with higher discrimination performance. A complex discriminator is used to determine whether a face is present. It is fast because there is no need to make complex judgments for all candidates.

  On the other hand, it has been proposed to improve image quality based on such a result of automatically recognizing a subject in an image. For example, Patent Document 1 discloses a method of automatically detecting a human face from an image and correcting the brightness of the image based on the result.

JP-A-6-303536 JP 2005-4287 A

  However, the face detection result is not necessarily 100% successful. Therefore, in the technique of Patent Document 1, for example, when a false detection of a relatively dark pattern occurs in a human photograph taken with appropriate brightness, the human face portion with appropriate brightness becomes even brighter. The image quality will be worse than the original image.

  In practice, such automatic object recognition processing required for image quality improvement processing can effectively improve image quality within a range where the image quality does not deteriorate. For example, in the correction of image brightness, a performance with a relatively high detection rate for a bright face and a relatively low false detection rate for a dark face is required. For this purpose, it is necessary to control the subject detection performance in accordance with the brightness of the subject to be detected. As a method for controlling the subject detection performance as described above, Patent Document 2 increases the detection rate of a dark face by changing a threshold according to the brightness of a pattern. However, since adverse effects due to erroneous detection of a dark face are not noticed, the occurrence of side effects is unavoidable when trying to correct the brightness. Moreover, it is not possible to suppress erroneous detection with high accuracy by simple threshold control.

  SUMMARY OF THE INVENTION In view of the above-described problems, an object of the present invention is to enable highly accurate subject detection suitable for applying image quality improvement processing.

  The image processing apparatus of the present invention is an image processing apparatus that detects a predetermined subject pattern from an image, and includes a collation pattern extraction unit that extracts a predetermined partial region for collation with the subject pattern from the image, and the collation pattern First pattern determining means for determining whether or not the partial area extracted by the extracting means is the subject pattern, and an image of the partial area determined to be the subject pattern by the first pattern determining means Evaluation means for evaluating characteristics and second pattern discrimination means for discriminating whether or not the partial area is the subject pattern based on an evaluation result by the evaluation means.

  According to the present invention, highly accurate subject detection suitable for applying image quality improvement processing can be performed. Thereby, it is possible to suppress the occurrence of side effects in the image quality improvement processing.

It is a block diagram which shows the structural example of the image processing apparatus of embodiment of this invention. It is a flowchart which shows an example of the process sequence of face recognition. It is a figure which shows a mode that the partial area | region of a predetermined magnitude | size is extracted from a reduced image. It is a figure which shows an example of the collation pattern of a face area. It is a figure which shows what combined the local area | region of various shapes and sizes. It is a flowchart which shows the procedure which combines the characteristic effective for the discrimination | determination of a to-be-photographed object. It is a block diagram which shows the detailed structural example of a 1st face discrimination | determination part. It is a block diagram which shows the detailed structural example of a weak discriminator. It is a block diagram which shows the structural example of the combination discriminator connected in series.

(First embodiment)
Hereinafter, the present invention will be described in detail according to preferred embodiments with reference to the accompanying drawings.
FIG. 1 is a block diagram illustrating a configuration example of an image processing apparatus according to the present embodiment.
In FIG. 1, reference numeral 10 denotes an image input unit, which is composed of, for example, a digital still camera, a camcorder, a film scanner, etc., and inputs image data by imaging or other known means. Further, it may be an interface device of a computer system that reads image data from a storage medium that holds digital image data.

  An image memory 20 temporarily stores the image data output from the image input unit 10. An image reduction unit 30 reduces the image data stored in the image memory 20 according to a predetermined magnification and outputs the reduced image data. Reference numeral 40 denotes a collation pattern extraction unit, which extracts a predetermined partial area from the image data reduced by the image reduction unit 30 as a pattern to be collated.

  Reference numeral 50 denotes a luminance correction unit that corrects the luminance distribution of the collation pattern extracted by the collation pattern extraction unit 40. Reference numeral 60 denotes a first face discriminating unit that discriminates whether the collation pattern extracted by the collation pattern extraction unit 40 and corrected by the luminance correction unit 50 is a face pattern or a non-face pattern. Reference numeral 70 denotes a brightness evaluation unit, which receives an image pattern from the matching pattern extraction unit 40 before performing luminance correction of the matching pattern determined by the first face determination unit as a face pattern, and evaluates the brightness.

  Reference numeral 80 denotes a second face discriminating unit that discriminates whether the collation pattern determined to be darker than the predetermined brightness by the brightness evaluation unit 70 is a face pattern or a non-face pattern. Reference numeral 90 denotes a face area output unit which outputs a pattern area determined as a face by the first face determination unit 60 or the second face determination unit 80. Reference numeral 100 denotes an image quality improvement unit that corrects the brightness of the image data input by the image input unit 10 based on the brightness of the pattern output from the face area output unit 90. The operation of each of the above blocks is controlled by a control unit (not shown).

Next, the operation of this embodiment will be described with reference to FIG. FIG. 2 is a flowchart showing an example of a face recognition processing procedure in the present embodiment.
First, in step S <b> 101, the image input unit 10 reads desired image data into the image memory 20.

  The image data read here is, for example, two-dimensional array data composed of 8-bit pixels, and is composed of R, G, B, and three surfaces. At this time, if the image data is compressed by a method such as JPEG, the image data is decompressed according to a predetermined decompression method to obtain image data composed of RGB pixels. Furthermore, in this embodiment, RGB data is converted into luminance data, and the luminance image data is applied to the subsequent processing and stored in the image memory 20. When YCrCb data is input as image data, the Y component may be directly used as luminance data.

  Next, in step S102, the image reduction unit 30 reads the luminance image data from the image memory 20, and generates luminance image data reduced to a predetermined magnification. In the present embodiment, detection is sequentially performed on image data of a plurality of sizes in order to support detection of faces of various sizes by the method described in Patent Document 2. For example, reduction processing to a plurality of images with different magnifications by about 1.2 is sequentially applied for subsequent detection processing.

  Next, in step S103, the matching pattern extraction unit 40 extracts a partial area having a predetermined size from the reduced luminance image data. FIG. 3 is a diagram illustrating a state in which a partial region having a predetermined size is extracted from the reduced image. The row A in FIG. 3 shows the respective reduced images reduced by the image reduction unit 30. In the present embodiment, a rectangular area having a predetermined size is cut out from each reduced image. FIG. 3B shows a state of clipping while repeating scanning from each reduced image in the vertical and horizontal directions. As shown in FIG. 3, when a face is determined by cutting out a matching pattern from an image with a large reduction ratio, a large face is detected for the image.

  Next, in step S104, the luminance correction unit 50 normalizes the luminance of the partial area cut out by the matching pattern extraction unit 40 based on the distribution. For example, brightness correction such as histogram smoothing is performed. This is for suppressing deterioration in accuracy of subject collation because the luminance distribution of the subject pattern to be captured changes depending on the illumination condition.

  Next, in step S105, the first face discriminating unit 60 discriminates whether or not the collation pattern extracted by the collation pattern extraction unit 40 and corrected by the luminance correction unit 50 is a face pattern or a non-face pattern. Detect patterns.

  Next, in step S <b> 106, the brightness evaluation unit 70 determines the brightness of the collation pattern determined as the face pattern by the first face determination unit 60. Specifically, the original image pattern before luminance correction corresponding to the matching pattern used for the determination is received from the matching pattern extraction unit 40, and the luminance average value is calculated. If the luminance average value calculated as the evaluation result is larger than the predetermined threshold as a result of this determination, it is determined that the face pattern is bright and the process proceeds to step S108. In other cases, it is determined that the face pattern is dark, and the process proceeds to step S107 to reduce false detection.

  Next, in step S107, the second face discriminating unit 80 determines that the first face discriminating unit 60 determines that the face pattern is a face pattern, and the brightness evaluation unit 70 determines that the face pattern is a dark face pattern. Determine whether the pattern or non-face pattern. Then, a face pattern is detected.

  In step S108, the face area output unit 90 outputs the rectangle extracted by the matching pattern extraction unit 40 to the face area as a result of detection by the first face determination unit 60 or the second face determination unit 80. Output as.

  Next, in step S109, it is determined whether scanning has been completed for all regions and magnifications. As a result of the determination, if scanning is completed for all regions and magnifications, the process proceeds to step S111. On the other hand, if the result of determination in step S109 is not complete, it is determined in step S110 whether or not to change the magnification of the rectangle. As a result of this determination, if the magnification is not changed, the process returns to step S103, and if the magnification is changed, the process returns to step S102.

  As described above, in the processing from step S103 to step S108, the reduced luminance image output from the image reducing unit 30 is repeatedly scanned in the vertical and horizontal directions in predetermined steps as shown in FIG. In addition, reduction processes with different magnifications are sequentially applied, and the processes from step S102 to step S108 are repeated.

  In step S111, the image quality improvement unit 100 corrects the brightness of the image data input by the image input unit 10 based on the brightness of the pattern output from the face area output unit 90. Note that the brightness correction here is performed in such a way as to brighten an input image captured with a dark face due to backlight or the like.

The details of the face discrimination method in the first face discriminating section (first pattern discriminating means) 60 and the second face discriminating section 80 (second pattern discriminating means) will be described below.
When a person discriminates a subject, attention is paid to characteristics of various parts of the subject, and it is considered that features effective for discrimination of the subject are combined from those features. For example, when determining a face, the result of whether or not there is a feature similar to a part such as an eye, mouth, or nose is comprehensively determined to determine whether the target pattern is a face. . Therefore, in the present embodiment, a large number of local feature amounts are extracted as subject features, a combination discriminator having good performance is obtained by boosting learning using the local feature amounts as weak discriminators, and used for subject discrimination.

  In the present embodiment, a linear identification feature in a local region having a predetermined size and shape is used as a local feature amount. FIG. 4 is a diagram illustrating an example of a face area matching pattern. For example, if the collation pattern F is a luminance pattern composed of 20 × 20 pixels as shown in FIG. 4, it is a 5 × 5 pixel square local region P that is a part thereof. In this case, if similar local regions P are set at all positions in the collation pattern, 256 local regions can be considered.

And as a local feature-value, the feature extraction kernel based on linear discrimination is calculated | required about each local region using many samples of the pattern which is a face among the collation patterns, and a non-face (non-face) pattern. The most widely known linear discriminant method is Fisher's linear discriminant function, which calculates the mean and variance of face samples in the local region (μ _O , σ _O ² ) and the mean and variance of non-face samples (μ _N , When σ _N ² ), the feature extraction kernel φ is obtained by the following equation 1. Note that argmax means obtaining a parameter that maximizes the value in parentheses, and φ can be obtained by solving the eigen equation.
φ = argmax {(μ _O −μ _N ) ² / (σ _O ² + σ _N ² )} (Formula 1)

  As local feature amounts, as shown by P1, P2, and P3 in FIG. 5, various shapes such as horizontally long rectangles and vertically long rectangles and sizes obtained from local regions may be used in combination. In addition, the original matching pattern may be reduced to create various low-resolution matching patterns, and the local feature amounts in the matching patterns of each resolution may be used in combination.

  In this embodiment, the feature extraction kernel is derived from Fisher's linear discriminant function, but the linear discriminant function may be obtained by a linear support vector machine. Further, MRC may be used. Further, the present invention is not limited to the linear discriminant function, and for example, a non-linear discriminant function constituted by a neural network may be used.

  Next, a description will be given of a method for combining features effective for subject discrimination from the above local feature amounts. In this method, the feature with the best discrimination performance is first extracted from the discrimination feature quantity, and another feature is added to complement the performance. It is an algorithm based on the idea. FIG. 6 is a flowchart showing an example of a procedure for combining features effective for subject determination from local feature amounts.

First, in step S201, a sample necessary for learning is input. In this embodiment, face and non-face matching patterns are used as samples. That is, if the sample feature vector is x _i , x _i is a 400-dimensional feature vector composed of 400 luminance data when the collation pattern is a luminance pattern of 20 × 20 pixels. Further, the class to which the sample belongs is y _i, and y _i represents whether it is a face, and takes a value of 1 for a face and −1 for a non-face. Here, i represents a sample number, and i = 1, 2,... N, where N is the total number of samples.

In step S202, the sample weights are initialized to be equal. That is, the sample weight is d _i and d _i = 1 / N is initialized.

Next, in step S203, local features are extracted from the samples and their weights. Note that the local feature obtained here is φ obtained by linear identification for each predetermined local region of the matching pattern as described above. However, a plurality of feature extraction kernels φ _j (j = 1, 2,... M, where M is the number of defined regions in the local region) are extracted by defining the local feature region. The average and variance of the face and non-face samples used when obtaining the feature extraction kernel use weighted values in consideration of the sample weight.

Next, in step S204, the discrimination performance of each local feature is evaluated using a sample, and the feature with the best discrimination performance is extracted as a weak discriminator. That is, first, using the feature extraction kernel φ _j of each local feature, a feature amount u _ij is calculated from the local pattern (referred to as z _i ) corresponding to each local region of the sample x _i as shown in the following Expression 2 ( T represents the transpose of the vector).
u _ij = φ _j ^T z _i (Formula 2)

Then, the sample distribution of the feature amount is estimated from a histogram having a bin having a predetermined size. When the value of the k th bin of the histogram obtained from the face sample is W ₊ ^k and the value of the k th bin of the histogram obtained from the non-face sample is W ₋ ^k , The local feature φ _t that minimizes the value of is extracted from the M local features. Note that the value of the histogram obtained here is the total sum of the sample weights d _i within the bin range corresponding to the feature quantity u _ij .
φ _t = argmin {Σ _k (√W ₊ ^k · W ₋ ^k )} (Expression 3)

That is, φ _t is a feature extraction kernel of the first weak discriminator t = 1, and the output of the weak discriminator represents the reliability of the subject to be discriminated and is a probability density (a face corresponding to each value of the feature amount ( Frequency) and the non-face probability density (frequency). That is, the value of Equation 4 below is used as a table for the bin size. However, ε is an appropriate small constant for avoiding divergence.
h _t = 1/2 · ln {(W ₊ ^k + ε) / (W ₋ ^k + ε)} (Expression 4)

Instead of the expression 4, the value of the following expression 5 may be output.
h _t = (W ₊ ^k− W ₋ ^k + ε) / (W ₋ ^k + W ₋ ^k + ε) (Expression 5)

Next, in step S205, the sample weight is updated based on the value of each sample output value h _ti in the extracted weak _classifier . Here, the update formula follows Formula 6 below.
d _i = d _i × exp {−y _i · h _ti } (Formula 6)

  Note that the weight of the sample that is erroneously determined by the weak classifier extracted by Equation 6 is large, and the weight of the sample that is correctly determined is updated to be small.

Next, in step S206, the processes in steps S203 to S205 are repeated a predetermined number of times, and it is determined whether or not a predetermined number of weak classifiers have been obtained. As a result of the determination, if the predetermined number of times has not been repeated, the process returns to step S203. On the other hand, if the result of determination in step S206 is repeated a predetermined number of times, the process proceeds to step S207. Then, a classifier combining the obtained weak classifiers is output. That is, when combining T weak discriminators, the output H of the combination discriminator is as shown in Equation 7 below.
_{H = Σ t h t (t} = 1,2, ··· T) ··· ( Equation 7)

  The feature extraction kernel of each weak discriminator necessary for obtaining the output H of the combination discriminator, the coordinate value specifying the local region, and the reliability for converting the feature quantity obtained by the feature extraction kernel into the output of the weak discriminator Save the table as a learning result.

  A method for performing subject discrimination using the above learning algorithm will be described below. That is, the content described below is the processing content of the first face discriminating unit 60 and the second face discriminating unit 80 in FIG. Hereinafter, the first face determination unit 60 will be described as an example.

FIG. 7 is a block diagram illustrating a detailed configuration example of the first face determination unit 60. FIG. 8 is a block diagram showing a detailed configuration example of each weak classifier.
First, from the collation pattern input to the first face discriminating unit 60, the local region extraction unit 6011 of the first weak discriminator 601 extracts the luminance pattern z of the local region necessary for calculating the feature amount. Then, the feature amount calculation unit 6013 calculates the feature amount u ₁ from the feature extraction kernel φ ₁ obtained by learning stored in the kernel storage unit 6012 as shown in the following Expression 8.
u ₁ = φ ₁ ^T z (Formula 8)

Then, the table conversion unit 6015 refers to the data in the reliability table storage unit 6014 and converts the subject reliability value to the first weak discriminator by table conversion based on the bin number to which the obtained feature quantity u ₁ belongs. 601 output. Similarly, the output of the T-th weak discriminator 60T is obtained from the second weak discriminator 602, the sum is obtained by the adder 6001, and is used as the output of the combination discriminator. Finally, the threshold processing unit 6002 determines whether the input pattern of the first face determination unit 60 is a face based on a predetermined threshold, and outputs the result.

  It is realistic in terms of processing speed to connect the combination classifiers obtained by learning in series to form a cascade type face detector. According to this, the configuration of the first face discriminating unit 60 is as shown in FIG. 9 instead of that shown in FIG. That is, the first combination discriminator 61, the second combination discriminator 62,..., The K combination discriminator 6K are connected in series. Note that each combination discriminator has a weak discriminator combination as shown in FIG.

  Note that the second face discriminating unit 80 performs learning as follows using the algorithm shown in FIG. 6 in order to accurately remove erroneous detection patterns in dark face patterns. First, as a sample, a face and non-face matching pattern determined as a face pattern by the first face determination unit 60 and determined as a dark face pattern by the brightness evaluation unit 70 is used. Similar to the first face discriminating unit 60, learning is performed by the procedure of steps S201 to S206, and the feature extraction kernel of the weak discriminator, the coordinate value specifying the local region, and the feature quantity obtained by the feature extraction kernel are obtained. A reliability table to be converted into the output of the weak classifier is obtained. That is, the second face discriminating unit 80 has the same configuration as shown in FIGS. 7 to 9 although it is necessary to prepare separately as a parameter for discriminating the face.

  As described above, the present embodiment is configured to accurately remove the erroneous detection pattern in the dark face pattern particularly for the dark pattern. Therefore, when performing the image quality improvement process for correcting the brightness of the image. The occurrence of side effects can be greatly suppressed.

  Of course, in the present embodiment, the first face discriminating unit 60 and the second face discriminating unit 80 are not limited to those described above, and the method described in Patent Document 2 or other face discriminating methods. You may comprise using. The first face discriminating unit 60 and the second face discriminating unit 80 do not necessarily need to use the same discrimination method. For example, the method used in the present embodiment may be used in the first face determination unit 60 and the method described in Patent Document 2 may be used in the second face determination unit 80.

  In the above embodiment, the case where the face is detected from the image and the brightness of the image is corrected based on the result has been described. However, the present invention can also be applied to other image quality improvement processing. For example, when the color of the face area detected from the image is suitably corrected, side effects can be avoided by keeping the false detection rate of the non-face pattern of an inappropriate color low. That is, it is possible to keep the probability that a person's face part of an appropriate color will be deteriorated to a bad color by color conversion due to erroneous detection. In this case, color is evaluated as an image characteristic for the pattern determined as a face by the first face determination unit.

  For example, the collation pattern is divided into regions based on the color information, and the largest color region can be estimated as the skin region, so the average color is calculated. Then, it is determined whether or not the Euclidean distance between the calculated average color and a predetermined color (in this case, the central color of the skin color that a person feels preferable) is greater than or equal to a predetermined threshold. Then, for a collation pattern having an average color equal to or greater than a threshold value, the second face discriminating unit learned using such a pattern is further discriminated.

  In the above embodiment, the human face is detected as the subject pattern, but other subject patterns may be used. Of course, the method of the present invention can also be realized by executing a program loaded in a computer.

(Other embodiments according to the present invention)
Each means constituting the image processing apparatus and each step of the image processing method in the embodiment of the present invention described above can be realized by operating a program stored in a RAM or ROM of a computer. This program and a computer-readable recording medium (storage medium) recording (storing) the program are included in the present invention.

  In addition, the present invention can be implemented as a system, apparatus, method, program, recording medium (storage medium), or the like, and can be applied to a system including a plurality of devices. It may also be applied to an apparatus consisting of a single device.

  In the present invention, a software program for realizing the functions of the above-described embodiments (in the embodiment, a program corresponding to the flowcharts shown in FIGS. 2 and 6) may be directly or remotely supplied to the system or apparatus. Including. This includes the case where the system or the computer of the apparatus is also achieved by reading and executing the supplied program code.

  Accordingly, since the functions of the present invention are implemented by computer, the program code installed in the computer also implements the present invention. In other words, the present invention includes a computer program itself for realizing the functional processing of the present invention.

  In that case, as long as it has the function of a program, it may be in the form of object code, a program executed by an interpreter, script data supplied to the OS, and the like.

  Examples of the recording medium (storage medium) for supplying the program include a flexible disk, a hard disk, an optical disk, and a magneto-optical disk. Further, there are MO, CD-ROM, CD-R, CD-RW, magnetic tape, nonvolatile memory card, ROM, DVD (DVD-ROM, DVD-R) and the like.

  As another program supply method, there is a method of connecting to a homepage on the Internet using a browser of a client computer. Further, the computer program itself of the present invention or a compressed file including an automatic installation function can be downloaded from the homepage by downloading it to a recording medium (storage medium) such as a hard disk.

  It can also be realized by dividing the program code constituting the program of the present invention into a plurality of files and downloading each file from a different homepage. That is, the present invention includes a WWW server that allows a plurality of users to download a program file for realizing the functional processing of the present invention on a computer.

  As another method, the program of the present invention is encrypted, stored in a recording medium (storage medium) such as a CD-ROM, distributed to users, and a homepage is established via the Internet for users who have cleared predetermined conditions. Download key information to decrypt from. It is also possible to execute the encrypted program by using the key information and install the program on a computer.

  Further, the functions of the above-described embodiments are realized by the computer executing the read program. Furthermore, based on the instructions of the program, an OS or the like running on the computer performs part or all of the actual processing, and the functions of the above-described embodiments can be realized by the processing.

  Furthermore, as another method, a program read from a recording medium (storage medium) is first written in a memory provided in a function expansion board inserted into the computer or a function expansion unit connected to the computer. Then, based on the instructions of the program, the CPU or the like provided in the function expansion board or function expansion unit performs part or all of the actual processing, and the functions of the above-described embodiments are also realized by the processing.

40 collation pattern extraction unit, 60 first face discrimination unit, 70 brightness evaluation unit, 80 second face discrimination unit

Claims (7)

An image processing apparatus for detecting a predetermined subject pattern from an image,
Collation pattern extraction means for extracting a predetermined partial region for collation with the subject pattern from an image;
First pattern determining means for determining whether or not the partial area extracted by the matching pattern extracting means is the subject pattern;
Evaluation means for evaluating image characteristics of the partial area determined to be the subject pattern by the first pattern determination means;
An image processing apparatus comprising: a second pattern determining unit that determines whether the partial area is the subject pattern based on an evaluation result by the evaluating unit.   The image processing apparatus according to claim 1, wherein image characteristics of the image are converted based on a detection result of the subject pattern.   The image processing apparatus according to claim 1, wherein the image characteristic is image brightness.   The image processing apparatus according to claim 1, wherein the image characteristic is a color of an image.   The image processing apparatus according to claim 1, wherein the subject is a face. An image processing method for detecting a predetermined subject pattern from an image,
A matching pattern extraction step for extracting a predetermined partial region for matching with the subject pattern from an image;
A first pattern determining step for determining whether or not the partial area extracted by the matching pattern extracting step is the subject pattern;
An evaluation step for evaluating an image characteristic of the partial area determined to be the subject pattern by the first pattern determination step;
An image processing method comprising: a second pattern discrimination step for discriminating whether or not the partial area is the subject pattern based on an evaluation result in the evaluation step.   A program for causing a computer to execute each step of the image processing method according to claim 6.

Images