KR100726473B1 - Apparatus for classifying an image and method therefor - Google Patents

Abstract

Disclosed are an image type classification apparatus and a method thereof. First, the type of the input image is discriminated based on the similarity between the external feature of the input image and the external feature of various kinds of reference images, and whether the internal feature of the input image is similar to the internal feature of the image type of the primary discrimination result. Determine whether the type of the input image identified by the difference is decoded by itself. In this way, it is possible to prevent unnecessary decoding by identifying whether the input image is a kind of image that can be decoded by itself.

Description

Apparatus for classifying an image and method therefor}

1A to 1F illustrate an example of an image code;

2 is a view showing the configuration of an embodiment of an image classification apparatus according to the present invention;

3 is a diagram illustrating a detailed configuration of an image input unit of an image classification apparatus according to the present invention;

4A is a view showing a detailed configuration of an external feature grasping unit of the image classification apparatus according to the present invention;

4B is a view showing a detailed configuration of the internal feature grasping unit of the image classification apparatus according to the present invention;

5 is a flowchart showing a flow of an embodiment of an image code classification method according to the present invention;

6 is a flow chart showing the flow of one embodiment of an image classification method according to the present invention.

The present invention relates to an image classification apparatus and a method thereof, and more particularly, to classify where an image type corresponds to an image code, a logo, a photo, etc., and to perform decoding according to the classified type on its own or through an external server. An apparatus and a method thereof are provided.

With the rise of mobile computing environment, interest in interface technology that accesses a server on a computer or a network and receives a predetermined service through information extracted by recognizing, analyzing, and extracting images through a camera is increasing. Technology related to image recognition includes biometric technology for recognizing face, fingerprint, back vein, iris, gesture, etc., watermark technology for hiding and extracting a specific pattern on the image, and technology for recognizing logos and characters of surrounding objects. Etc.

As the performance of personal mobile devices such as mobile computers, mobile phones, PDAs, etc. is gradually improved, a large amount of information can be processed on the terminal itself, but since the image recognition is very complicated and computational, all processes are still performed on the terminal. It cannot be processed. Therefore, some image codes such as color codes and barcodes and simple face recognition can be processed in the terminal, but complex image codes, biometric information, logos, etc. are transmitted to the external server, not the terminal, to interpret the information.

When sending an image to an external server, the terminal must tolerate the relatively slow network speed and receive the decoded result back from the server. For this reason, it takes a few seconds to several minutes for users to send the captured image to the server and receive the processing result from the server. If the image quality is poor or the decoding fails, the image must be retaken and sent. However, there is a disadvantage in that a communication fee of a terminal for exchanging data with a server is generated.

Conventional image recognition services generally provide services by recognizing only certain types of images. For example, the biometric field recognizes only the corresponding biometric image, and the image code field recognizes only the image code image. Of course, the image code service recognizes one-dimensional barcodes, two-dimensional codes, etc., but there is no case in which a single terminal recognizes biometrics, code recognition, and log recognition. This is because it is not possible to distinguish whether an image input from one terminal is an image code, a biometric image, a logo or a letter.

The technical problem to be achieved by the present invention is to determine whether the image input through a predetermined image input device is a decodable image itself or not, and if it is impossible to decode itself image separation for transmitting only the characteristic information of the image to the outside An apparatus and a method thereof are provided.

Another object of the present invention is to provide a computer-readable recording medium having recorded thereon a program for executing the above image classification method on a computer.

In order to achieve the above technical problem, an embodiment of the image classification apparatus according to the present invention, an image input unit for receiving an analysis target image; An external feature determiner which first determines a type of the analysis target image based on a similarity between the external feature of the analysis target image and the external feature of various kinds of reference images; An internal feature identifying unit for secondly determining whether an internal feature of the analysis target image is similar to an internal feature of an image type of the first discrimination result; And an image determination unit to determine whether the type of the analysis target image determined based on the first and second determination results is a type that can be decoded by itself.

In order to achieve the above technical problem, an embodiment of the image classification method according to the present invention comprises: receiving an analysis target image; Firstly determining a type of the analysis target image based on a similarity between the external feature of the analysis target image and the external feature of various kinds of reference images; Secondly determining whether an internal feature of the analysis target image is similar to an internal feature of an image type of the first determination result; And determining whether the type of the analysis target image determined based on the first and second discrimination results is a type that can be decoded by itself.

As a result, it is possible to determine whether the input image is an image of a type that can be decoded by itself, thereby preventing unnecessary decoding.

Hereinafter, with reference to the accompanying drawings will be described in detail with respect to the image fractionation apparatus and method according to the present invention.

1A to 1F are diagrams illustrating an example of an image code.

Figure 1a is a data matrix, Figure 1b is a QR code, Figure 1c is a PDF-417 code, Figure 1d is a color code and gray code, Figure 1e is a mixed code generated by superimposing the color code and QR code, Figure 1f is a color code Is an embodiment of a mixed code generated by superimposing a predetermined image.

In an exemplary embodiment of the present invention, the classification of the image code shown in FIGS. 1A to 1F is disclosed, but other image codes may be defined and classified, and a logo or a photograph may be classified. In the present invention, for convenience of description, a method of classifying image codes will be mainly described.

2 is a diagram illustrating a configuration of an embodiment of an image classification apparatus according to the present invention.

2, the image classification apparatus includes an image input unit 200, an external feature determiner 210, an internal feature determiner 220, an image determiner 230, a decoder 240, and a feature information extractor ( 250).

The image input unit 200 receives an analysis target image through a predetermined image input device. The image input device is a camera built in or external to a personal portable terminal, or a device for receiving an analysis target image through a wired / wireless communication network in the form of an electronic document. The detailed configuration of the image input unit will be described in detail with reference to FIG. 3.

The external feature determiner 210 first determines the type of the analysis target image based on the outline shape and the pattern of the analysis target image. The external feature determiner 210 first determines the type by comparing information about external features and patterns of various types of images stored in advance and external features of the analysis target image. Hereinafter, various types of images that are analyzed and stored in advance are referred to as 'reference images'. For example, since the bar code is composed of a plurality of black bars, and the two-dimensional image is composed of a constant shape such as a square or a rectangle, the external feature acquiring unit 210 determines the type when the outer shape of the analysis target image is a constant shape. Firstly, it is recognized as a two-dimensional image. Detailed configuration of the external feature grasping unit 210 will be described in detail with reference to FIG. 4A.

The internal feature determiner 220 secondly determines the type of the analysis target image based on the internal pattern and the color of the analysis target image. Since the type of the analysis target image is first narrowed to a certain range by the external feature identification unit 210, the analysis target image is compared by comparing the internal pattern and color of the reference image within the range with the internal pattern and color of the analysis target image. Figure out the type of.

For example, in the case of a barcode, it does not include another pattern or color inside the black bar, so it is not necessary to identify an internal feature. However, in the case of a two-dimensional image code, only the outline shape indicates whether the image to be analyzed is a two-dimensional image code. Indistinguishable. Therefore, even if the shape or pattern of the outline is primarily the shape of the image code, it is necessary to know whether the internal pattern includes the features of the two-dimensional image code. Detailed configuration of the internal feature grasping unit 220 will be described in detail with reference to FIG. 4B.

The image discriminating unit 230 determines the type of the analysis target image based on the primary discrimination result and the secondary discrimination result, and determines whether the image is decodable by itself. However, even if the type of image to be analyzed in the primary determination result is determined by the two-dimensional image code, and the type of the image to be analyzed in the secondary determination result is determined by the color code among the two-dimensional image codes, the number of colors used in the color code is constant. If it is below, it is not a normal color code. Therefore, the image discriminator 230 determines whether the type of the analysis target image determined through the first and second discrimination results is decodable.

For example, the color code is composed of cells using four or eight colors or three or more contrast information. If the color code consists of only a single color, it is not a normal color code. 230) is classified as undeterminable despite the first and second discrimination results. Similarly, 2D codes are classified as undeterminable because the similarity becomes lower when a large part of the code image is damaged by reflection of light or ink. In these cases, the image is badly damaged and must be binarized again or inputted again. Even in the case of an indeterminate form, the similarity can be determined.

When the image to be analyzed is determined by the image determiner 230 as an image type that can be decoded by the image determiner 230, the decoder 240 decodes the image to be analyzed by a decoding method corresponding to the identified type. In the case of letters and numbers, the decoding unit 240 determines letter and number information using a pattern matching technique through feature information. In the case of the 2D image code and the color code, the decoder 240 may perform decoding by identifying and normalizing the positions of the cells using the feature information. The decoding unit 240 extracts the most similar image from the database through a matching technique even in the case of an indeterminate form and provides a corresponding service connected thereto. Extraction of the feature information will be described in detail with reference to FIGS. 4A and 4B.

If the type of the analysis target image is determined to be a color code, a two-dimensional code, a bar code, etc. having a low capacity, since the decoding is generally possible in consideration of the performance of the terminal, the decoding unit 240 decodes the analysis target image by itself. However, in the case of high-capacity image codes, there is a case in which the processing of the mobile terminal is impossible due to the limitation of the information processing capability of the portable terminal.

When the analysis target image is an image that cannot be processed by itself, the feature information transmitting unit 250 transmits the analysis target image to the high performance external server so as to decode the analysis target image and receives the result. Examples of images that cannot be self-processing include some of the irregular logos, complex pictorial images, and many letters and numbers. However, since these images can also be decoded by itself as the performance of the terminal equipped with the image classification apparatus according to the present invention is improved, the image types that cannot be processed by themselves are dynamically set according to the performance of the terminal.

The feature information transmitter 250 does not transmit the analysis target image itself to the external server in consideration of the communication speed with the external server, but analyzes the characteristic of the analysis target image and transmits only the analyzed feature information to the external server. desirable. Extraction of feature information for transmission will be described in detail with reference to FIGS. 4A and 4B.

3 is a diagram illustrating a detailed configuration of an image input unit of the image classification apparatus of FIG. 2 according to the present invention.

Referring to FIG. 3, the image input unit 200 includes an image normalization unit 300, a binarization unit 310, and an image extraction unit 320. The image input unit 200 generates an analysis target image by separating a portion (that is, image code, logo, text, etc.) displaying predetermined information from an image input through a camera or the like from a background.

The image normalization unit 300 first converts an image input through the camera into an image format that is easy to decode. For example, the image normalization unit 300 converts an image input in a format such as YUV24 or RGB24 into an RGB24 format. The image normalization unit 300 generates a normalized image by correcting the color, contrast, and the like of the image and aligning the direction and size of the image.

The color and contrast of an image input through a camera or the like are greatly influenced by the type of lighting, the shooting time (night or day), the printing medium on which the image is printed, and the performance of the camera sensor. Therefore, the image normalization unit 300 corrects the color and contrast of the image by measuring the variation of the color of the input image, the variation of the contrast, and their distribution characteristics.

Looking at the correction process in detail, the image normalization unit 300 first divides the image by a predetermined block unit and obtains the contrast and lighting characteristic values of each block. The image normalizer 300 compares the values obtained for each block and interpolates each other to obtain contrast information of the entire image. This method is generally for processing gradation contrast information, and can be corrected using the change ratio of the contrast.

Contrast information is more accurate when comparing colors of the same kind. Therefore, the image normalization unit 300 photographs a predetermined image in a predetermined standard lighting environment and analyzes the distribution of contrast and the like in advance. The image normalization unit 300 may correct the contrast of the image more easily and accurately by comparing the distribution of the contrast, etc., previously analyzed, with the intensity distribution of the input image.

That is, the image normalization unit 300 analyzes in advance the distribution ratio, maximum value, minimum value, and maximum, minimum, median value, and distribution ratio of each of the R, G, and B channels of specific colors under standard illumination. The image normalizer 300 corrects the color and contrast of the input image by comparing the R, G, and B channel distribution ratios of the image input in the general lighting environment with the values under the standard illumination analyzed in advance.

The image normalization unit 300 may use a color correction algorithm, such as a gray world assemption method, a retinex algorithm, or a gamut mapping algorithm, to generate a normalized image with contrast and color corrected.

The binarization unit 310 converts the normalized image of which color and contrast are corrected into a binarized image represented by two colors or two contrasts. In this embodiment, the conversion to a binarized image composed of black and white is described.

The binarization unit 310 first divides the normalized image by color and contrast. For example, the binarization unit 310 divides an image into several areas having the same color and contrast by determining that the same color and contrast are within the predetermined error when the difference between the color and the contrast between adjacent pixels is within a certain error.

When segmentation using color and contrast is completed, the binarization unit 310 obtains a binarization value through brightness and histograms of R, G, and B channels. The binarizer 310 obtains an average by grouping adjacent pixels in several units and weights the average, and sets the average as the frequency of the corresponding brightness. If the brightness value or the color value of the normalized image can be clearly distinguished into two groups, the binarization unit 310 selects the brightness value or the intensity of the color having the least frequency between the two groups as the binarization value. If the group is divided into several groups, the binarization unit 310 memorizes the brightness values and the color values having the least frequency among the groups as binary environment variables in advance, and selects one of them as a binarization value.

The binarization unit 310 binarizes the image to black and white using the set binarization value, and then removes miscellaneous images. For example, the binarization unit 310 converts the pixel to white when the brightness of the image pixel is greater than or equal to the binarization value, and converts the pixel to black when the binarization value is less than the binarization value.

If the size of the black object is less than or equal to the predetermined size in the image converted into black and white, the binarization unit 310 determines that it is a ghost and removes it. However, 2D black and white image code is composed of a set of small black objects, so they cannot be eliminated uniformly. Accordingly, the binarization unit 310 may protect the code image that is expected to be located in or near the image by first removing the outermost black pixels of the image, that is, the blurry images connected to the outer frame. If the image is not uniform, the binarization unit 310 restores the broken points of the image through Gaussian filtering.

The image extractor 320 detects an analysis target image area from the binarized image. First, the image extractor 320 detects a candidate region in which objects exist by finding an outline of a region shown in black in the binarized image. In the case of letters or numbers, an amorphous figure having a thin line or a relatively thin thickness is detected as a candidate region, and in the case of a color code or a logo, a region having a relatively predetermined shape is detected as a candidate region. In addition, in the case of a QR code, a data matrix, or the like, a candidate region consisting of a set of pixels and lines is detected.

The candidate region detected by the image extractor 320 may be one region or may consist of several regions. Therefore, the image extracting unit 320 processes the candidate region, which is separated from each other, into one analysis target image region. For example, since an image code such as a QR code is a set of several pixels, it is preferable to calculate the inter-pixel distance and to process pixels having a predetermined distance or less as one set. In particular, it is more useful to extend the range by calculating the distance from adjacent pixels based on the object located in the center of the image. This is because the object of interest is in the center of the image in most cases.

4A is a diagram illustrating a detailed configuration of an external feature identifying unit of the image classification apparatus according to the present invention. Referring to FIG. 4A, the external feature determiner 210 includes an external feature extractor 410 and a first pattern similarity calculator 420.

The external feature extractor 410 detects the outline of the image to be analyzed, the change in the directionality, the maximum and minimum points of the coordinates, and the length of the unidirectional segment to extract the feature information to generate standardized pattern information. The feature information includes the coordinates of the feature point, the chain code and the length between the feature points, and the like. Specifically, the feature information includes angle change information of an outline constituting an image or line segments constituting a feature pattern, a pattern based on coordinates and relative positions of feature points, maximum, minimum points, and pattern lengths of image external feature points.

The feature point means pixels representing the object's characteristics in an object such as a line segment or a figure on the image.As a method of finding the feature point, various algorithms such as an edge detection algorithm, an end of a character in character recognition, and an overlap detection method are combined and used. do. Commonly used feature points are main information of each pixel constituting a line segment of an object, for example, a terminal, a corner, a pixel in which the direction of the segment changes more than a limit angle, and a center point of 2D code cells. There is a reference about how to find these features.

"Ho-Geun Oh, Barcode Technology and Applications, Sung-angdang, 1997", "Seong-Hwan Lee, Character Recognition-Theory and Practice, Hongneung Science Publishing House, 1994", "Earl Gose, Richard Johnsonbaugh, Steve Jost, Pattern Recognition and Image Analysis, Prentice Hall, 1996 "," ISO / IEC 18004: 2000 Information techonolgy-Automatic identification and data capture technique-Bar code symbology specifications-PDF417 "," ISO / ISE 16022: 2000 Information technology-International symbology specification-Data Matrix "," ISO / IEC 16032 : 2000 Information technology-International symbology specification-MaxiCode ".

The external feature extractor 410 performs skeletalization, hook processing, smoothing, and normalization as needed to generate standardized pattern information. Explain each process in turn.

(1) skeletalization

Skeletalization refers to replacing line segments or letters with line segments of one pixel thickness in order to extract feature points of line segments or letters. Skeletalization is performed on objects that are below a certain thickness but can also be performed on thick objects as needed. When skeletalization is performed on a thick object, the line segment obtained through skeletalization includes information on the central axis of the object, so that the shape or characteristic of the object is well represented. For example, in the case of a square, skeletalization may yield an X-shaped skeletal segment that diagonally connects the relative vertices. This skeletal process is performed by a thinning algorithm.

(2) hook processing

Hook processing removes the squeaking of the feature point. For example, in the case of letters, there is a squeak that can be seen as a kind of miscellaneous at the beginning of the letter, which is preferably removed. The squeezing portion has a short distance between the feature points and a sharp change in direction, and is thus removed based on this feature.

(3) smoothing

Smoothing is to remove the skewness caused by the distortion of line segments. That is, in the case of a set of curves forming a line segment as a whole, although the direction change is somewhat, the feature points are dense and change direction, so it is smoothed to process them in a straight line or a curve.

For example, Ellozy's method (Equation 1), which makes lines flat when the letters or drawings are crooked, and Plamo's method of linearizing the points of locally straight parts as a whole (Equation 1) 2) etc.

Where x is a vector representing coordinates.

(4) normalization

Normalization means that the size and the direction of the image input through the camera may vary according to the characteristics of the camera, so that the normalization and the direction are normalized according to a predetermined criterion.

First, for size normalization, the length of the shortest line segment or n-divided line segment among the components of an object is set as the unit length, and the lengths of other line segments are identified as multiples of the unit length. Thereby, the length ratio of each line segment can be known. By using the length ratio of the line segment, the size of the image may be normalized by enlarging or reducing the length of the line segments constituting the outline of the object at the same ratio. That is, when the sum of the lengths of the line segments constituting the outline is divided by the unit length, the number of unit lengths is known, and thus the size of the image may be standardized to have the same unit length number.

Next, directional information, that is, chain code information, between predetermined feature points is configured for direction normalization. For example, the directional information from one reference point A connected to the reference feature point is obtained, and the directional information from another feature point B connected to A is obtained. In this way, if the direction information of all the line segments forming the outline is obtained, a continuous chain code is created. If the object is a shape rather than a line segment, the start feature will be the end feature, otherwise the start feature and the end feature will be different. When the chain cord is formed, the shape number is obtained using the difference in the direction numbers constituting the chain cord (for details, see "Digital Image Processing, by Rafael C. Gonzalez and Richard E. Woods, Addison-Wesley, 2002". ).

When determining the shape number, it is preferable to use an object that is not rotated, and when it is rotated, the degree is calculated in the clockwise direction on the basis of the smallest degree using the 8-way code technique. The obtained order is called the number of image pattern shapes.

The first pattern similarity calculation unit 420 compares the pattern of the feature information extracted from the analysis target image with predetermined reference patterns to find the most similar reference pattern. Here, the reference patterns are patterns (or number of pattern shapes) which are obtained by analyzing, in advance, reference images of various types of image codes or logos which can be processed by the personal portable terminal in which the image classification apparatus according to the present invention is installed. Accordingly, the first pattern similarity calculator 420 may primarily determine the type of the analysis target image by comparing the pattern (or pattern shape number) generated from the feature information of the analysis target image with the reference pattern (or pattern shape number). Can be.

In detail, the first pattern similarity calculator 420 obtains a reference pattern set having the same element by using the unit length number in the salping normalization process, that is, the number of elements of the set of shape numbers, and then, Similar reference patterns are found by comparing the element order of the number of shapes of the reference pattern. At this time, even if the minimum order, i.e., normalized to 0, there may be an image pattern type number starting with another minimum order 0, it should be compared in consideration of this. For example, when there is an image pattern form number "03313303", when the form number of the reference pattern is "03033133", both are the same image pattern.

If the analysis target image is a shape filled inside, a feature of one outline shape is extracted, and thus, the first pattern similarity calculator 420 may calculate the pattern similarity between the pattern of the analysis target image and the reference image pattern only once. However, if the image to be analyzed is a set of small objects with several outlines, a set of pattern similarities corresponding to one object should be obtained.

In the case of 2D image code, the similarity of the patterns constituting the data occurs, and the similarity of the Finder Pattern or the outlines to find the code. In the case of a QR code, there are three Finder patterns including three vertices of the code, and in the other parts, a set of patterns formed by various types of pixels are formed.

For example, the QR code of the 2D image code has a finder pattern in the remaining portion except the lower right side of the code, and the maxicode has a concentric finder pattern called Bull's eye in the center. In the case of a barcode, there are characteristic patterns which are guided on the left and right sides of the code. In addition to these pinter patterns, guide bars, other patterns that may characterize the code, such as timing patterns, alignment patterns, etc., are present at predetermined positions. In this case, one code is composed of a set of distributed patterns, that is, a set of outlines or points, so that the similarity of the corresponding codes can be measured comprehensively by calculating the similarity of each distributed pattern. In the case of a logo, it is necessary to calculate the similarity of these distributed patterns, and to calculate the similarity comprehensively as one logo using a set of such distributed patterns. It is most common to consider the relative distance between the dispersion patterns whether the dispersion patterns belong to one image, and this is commonly called a spiral search.

In FIG. 4A, a method of generating a normalized feature pattern shape number and comparing the reference pattern shape number is described. However, the present invention is not limited thereto, and the external feature of the analysis target image and the external feature of the reference image are compared with each other. However, it is characteristic to primarily determine the type of image to be analyzed, and various other methods for comparing external features may be used to accomplish this purpose.

Hereinafter, a specific example of first determining the type of the image by comparing the similarity between the feature of the analysis target image extracted by the external feature extractor 410 and the feature of the reference image (color code, barcode, etc.) will be described.

(1) classification of color codes

In the case of a color code filled with an area, since a square or a rectangle is in the form of an outline of an object, it is discriminated based on these characteristics.

(2) classification of barcode image codes

Since the barcode consists of a predetermined guide pattern and a set of white / black bars, it is discriminated based on the set of these patterns.

(3) classification of 2D image code

The 2D image code is largely divided into a matrix and a hierarchical structure, but most of them are composed of finder patterns and pixels of a data area. Finder Patterns typically exist at the vertices or outlines of code, and certain codes may be central. Finder Patterns are squares, certain black and white bar patterns or concentric circles, concentric squares, and outline shapes. Therefore, if there is such a pattern of the finder pattern in a predetermined position of the analysis target image, it is determined as sortable. The remaining parts other than the finder pattern are formed in an irregular pixel pattern, so there is no need to look at each one, but the overall shape of the object is made up of squares, rectangles, and predetermined shapes.

(4) other figures and line segments

Even if the shape of the object does not correspond to the above code shape, it is determined that the object can be classified if the shape of the overall shape of the specific figure (square, rectangle, octagon, hexagon, etc.) or similar to the already registered reference patterns. In general, this type of shape corresponds to the case of a predetermined image code or company logo. If the pattern is a set of relatively short straight and curved line segments, the similarity is measured by comparing the set of reference patterns corresponding to letters and numbers. In the case of an indeterminate form, the similarity is measured by comparing the set of reference patterns corresponding to the indeterminate form. If the degree of similarity is equal to or greater than the predetermined degree of similarity, it is determined as sortable.

(5) other

When the feature pattern of the image to be analyzed is less than a predetermined similarity when compared with the set of reference patterns, it is classified as impossible to classify, and the calculated binarization value and the characteristic information of the feature points are analyzed and used as reference data. That is, if the area of the object is small after the binarization and there are many disconnected line segments, the binarization value is so low that the portion to be treated as black is treated as white so that the main information of the object is removed. In the opposite case, it is necessary to increase the binarization value, so in this case, it is necessary to select a value suitable for the condition from the candidate binarization values previously determined.

4B is a diagram illustrating a detailed configuration of an internal feature grasping unit of the image classification apparatus according to the present invention. Referring to FIG. 4B, the internal feature determiner 220 includes an internal feature extractor 430 and a second pattern similarity calculator 440.

The internal feature extractor 430 detects feature points and color information inside the analysis target image determined to be primarily classified by the external shape of the image. Extraction of the internal feature may use various methods used by the external feature extractor 410. In the case of a barcode, since the outline is the outline of the patterns forming the object, it does not need to be separately obtained. However, in the case of two-dimensional code, another pattern exists inside the Finder Pattern.

For example, in the case of the Finder Pattern of the maxicode, since there are other concentric circles in the found outline, the internal feature extraction unit 430 finds the feature points of the concentric circles. Similarly, in the case of a QR code, since a black square exists in the Finder Pattern, the internal feature extraction unit 430 detects an internal feature of the black square. In the case of a color code, since various shapes exist in the outline, the internal feature extractor 430 detects color information in the color code as an internal feature as well as feature points for various shapes. In the case of the logo image, since the letters and numbers are often included in the figure, the internal feature extraction unit 430 determines the feature points for the letters or numbers, and determines the color of each area.

The second pattern similarity calculator 440 calculates a similarity between the internal feature point distribution and the reference patterns. That is, the second pattern similarity calculator 440 measures the similarity between the feature point, the feature information, and the reference patterns by the patterns inside the object found by the internal feature extractor 430. Since the image to be analyzed is first classified by the first pattern similarity calculator 420 to what kind of image, that is, to which candidate group, based on an external feature, the second pattern similarity calculator 440 is 1; Re-determine whether the difference is correct. Accordingly, the second pattern similarity calculator 440 determines the similarity very high if the internal feature points and the feature information of the analysis target image are the reference feature information belonging to the candidate group, and otherwise determines the similarity.

5 is a flowchart illustrating the flow of an embodiment of an image code classification method according to the present invention.

Referring to FIG. 5, first, an analysis target image is input (S500). First, it is determined whether the external feature of the image to be analyzed is similar to the external feature of the image code (S505). If the external feature of the image to be analyzed is primarily determined to have an external feature similar to the image code (S510), the second feature determines whether the internal feature of the image to be analyzed is similar to the internal feature of the image code (S515). The comparison between the image to be analyzed and the internal and external features of the image code uses the same method as the method used in FIGS. 4A and 4B.

If it is determined that the internal feature of the analysis target image is similar to the internal feature of the image code (S520), the analysis target image is decoded using the decoding method of the corresponding image code (S525). If the analysis target image is an image of a kind that cannot be processed by itself (that is, a logo or a character), the analysis target image is transmitted to an external server and the decoding result is received (S530). A corresponding service is provided based on the decoding result (S535).

6 is a flow chart showing the flow of one embodiment of an image classification method according to the present invention.

Referring to FIG. 6, an image printed on a predetermined print medium is received through an image input device (S600). A normalized image is generated by correcting distortion of color and contrast due to the surrounding environment at the time of image input (S605). The normalized image is converted into a binarized image having two colors of black and white (S610), and an analysis target image is extracted from the binarized image (S615).

The feature pattern is extracted by analyzing the shape and feature points of the outline of the analysis target image (S620), and the pattern similarity with the feature pattern (that is, the reference pattern) of the reference images is calculated (S625). If the pattern similarity is greater than or equal to a predetermined threshold, the analysis object image is first determined to be the same image type as the type of the predetermined reference image (S630), and an internal feature of the analysis object image is detected (S635). In operation S640, the internal feature of the analysis target image and the internal feature of the image type determined by the first determination result are similar. If the first and second determination results that the image type is an image type that can be decoded by itself (S650), decoding is performed by selecting a decoding method corresponding to the image type of the analysis target image (S655). When the type of the analysis target image identified as the first determination result and the second determination result is a type of an image that cannot be decoded by itself, the analysis target image is transmitted to the external server and the decoded result is received from the external server (S660). After decoding, the corresponding service is provided (S665).

The invention can also be embodied as computer readable code on a computer readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, and the like, and may also be implemented in the form of a carrier wave (for example, transmission over the Internet). Include. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

So far I looked at the center of the preferred embodiment for the present invention. Those skilled in the art will appreciate that the present invention can be implemented in a modified form without departing from the essential features of the present invention. Therefore, the disclosed embodiments should be considered in descriptive sense only and not for purposes of limitation. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the scope will be construed as being included in the present invention.

According to the present invention, an image may be classified into an image code, a logo, a picture, or the like through image processing on the input image, and decoding may be performed according to each type. Therefore, it is possible to quickly determine whether an image can be processed by itself, thereby preventing unnecessary decoding, and by extracting basic feature information in advance, image recognition can be processed more quickly when the terminal or server is decoding.

In addition, when recognizing a complex image or an image requiring many operations, it is possible to determine in advance whether to decode by a terminal or a server, thereby improving speed and accuracy. In addition, when the server decodes, unnecessary waste of time occurs because the conventional transmission is performed without determining whether or not the image is decoded in the terminal. However, according to the present invention, since only the decodable image is transmitted by the server, effective information processing is possible.

Claims (13)

An image input unit configured to receive an analysis target image; An external feature determiner which first determines a type of the analysis target image based on a similarity between the external feature of the analysis target image and the external feature of various kinds of reference images; An internal feature identifying unit for secondly determining whether an internal feature of the analysis target image is similar to an internal feature of an image type of the first discrimination result; And And an image discriminating unit configured to determine whether the type of the analysis target image determined based on the first and second discrimination results is a type that can be decoded by itself. The method of claim 1, A decoding unit which decodes the analysis target image by a decoding method corresponding to the image type, when the image type determined by the image discriminating unit corresponds to an image type that can be decoded by itself; And And a feature information transmitter for transmitting the analysis target image to an external server and decoding the image if the image type determined by the image discriminator corresponds to an image type that cannot be decoded by itself. . The method of claim 2, The internal feature grasping unit generates external feature information including information on feature points forming an outline of the analysis target image by performing skeletonization, hook processing, smoothing, and normalization on the analysis target image, The external feature identifying unit generates internal feature information including information and color information of internal feature points of the analysis target image. The feature information transmitter may transmit the external feature information and the internal feature information to the external server. The method of claim 1, wherein the image input unit, An image normalization unit configured to generate a normalized image by correcting variations in color and contrast according to an image input environment including lighting and a type of a print medium when the image is input through the image input device; A binarization unit configured to set a predetermined brightness as a threshold value and change a pixel brighter than the threshold value into black and a pixel darker than the threshold value into white in the normalized image to generate a binarized image; And And an image extractor configured to extract the area formed of black to generate the analysis target image. The method of claim 1, wherein the external feature grasping unit, After normalizing the size and direction of the analysis target image, constructing a chain code composed of direction information of line segments connecting the feature points constituting the outline of the analysis target image, and using the difference in the number of directions constituting the chain code An external feature extracting unit obtaining a pattern shape number; And And a first pattern similarity calculator configured to first determine the type of the analysis target image by comparing the similarity between the pattern shape number of the analysis target image and the pattern shape number of the reference images. The method of claim 5, The external feature extraction unit performs at least one of skeletalization, hook processing, and smoothing of the analysis target image. The method of claim 1, wherein the internal feature grasping unit, An internal feature extraction unit for extracting feature points and color information of the internal shape of the analysis target image; And And a second pattern similarity calculator configured to second determine whether the inner patterns represented by the feature points are similar to an inner pattern of the image type of the first determination result. 2. Receiving an image to be analyzed; Firstly determining a type of the analysis target image based on a similarity between the external feature of the analysis target image and the external feature of various kinds of reference images; Secondly determining whether an internal feature of the analysis target image is similar to an internal feature of an image type of the first determination result; And And determining whether the type of the analysis target image determined based on the first determination and the second determination results is a type that can be decoded by itself. The method of claim 8, Decoding the analysis target image by a decoding method corresponding to the image type when the type of the analysis target image corresponds to a type of an image which can be decoded therein; And And transmitting the analysis target image to an external server if the type of the analysis target image corresponds to an image type that cannot be decoded by itself. The method of claim 8, The primary determining step may include generating external feature information including information on feature points forming an outline of the analysis target image by performing skeletonization, hook processing, smoothing, and normalization on the analysis target image. and, The second determination step may include generating internal feature information including information and color information of internal feature points of the analysis target image. The external transmission of the analysis target image may include: transmitting the external feature information and the internal feature information to the external server when the classification target type corresponds to an image type that cannot be decoded by itself. Fractionation device. The method of claim 8, wherein the first determination step, After normalizing the size and direction of the analysis target image, constructing a chain code composed of direction information of line segments connecting the feature points constituting the outline of the analysis target image, and using the difference in the number of directions constituting the chain code Obtaining a pattern shape number; And And judging the type of the analysis target image by comparing the similarity between the number of pattern shapes of the analysis target image and the number of pattern shapes of the reference images. The method of claim 8, wherein the secondary determination step, Extracting feature points and color information of an internal shape of the analysis target image; And And secondly determining whether an inner pattern represented by the feature points is similar to an inner pattern of an image type of the first discrimination result. 2. A computer-readable recording medium having recorded thereon a program for executing the method according to any one of claims 8 to 12.

Images