JP2007310466A - Image tilt detection device and image tilt detection processing program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image tilt detection device capable of detecting the tilt of an image without requiring a large amount of a processing time by simplified processing and with high dimensional accuracy. <P>SOLUTION: In a wavelet transformation part 1, multilevel wave let space conversion is performed to the image of a tilt detection object. An angle detection part 2 performs Hough transformation to at least an LL block of multiple hierarchies, creates two-dimensional histogram for a ρ-θ plane, obtains an angle θ from a point of a maximum frequency. Sequentially from a deep hierarchy side, a candidate range is set up based on the angle obtained in a lower hierarchy, the Hough transformation is performed in its candidate range, the angle is gradually obtained with a higher degree of the accuracy. The LL block with the wavelet space conversion performed is lower in resolution than that of an original image, and high speed processing is achieved. Moreover, because noise, a halftone dot and the like are separated into other high-frequency wave blocks, the angle can be detected without being influenced by these. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a technique for detecting the inclination of an image.

  Many techniques for detecting the tilt of an image have been devised and used in practice. Typical examples include the following conventional techniques. First, the tilt detection technique described in Patent Document 1 rotates a document image by θ while sequentially changing the angle θ, creates a circumscribed rectangle including all black pixels included in the rotated image, The angle θ at which the area is minimized is used as the inclination angle of the image. The technique described in Patent Document 2 creates a circumscribed rectangle from the connectivity of black pixels. Histograms that are projected in various directions on one vertex of each circumscribed rectangle are obtained, and the angle with the maximum frequency is used as the skew angle.

  As another method, for example, in the method described in Patent Document 3, the image data is subjected to edge processing and then binarized, the binary image is subjected to Hough transform, and the parameter space obtained by the Hough transform (ρ A coordinate group whose frequency is equal to or higher than a predetermined value on the -θ plane) is generated, and the angle of the image is estimated from the coordinate group.

JP-A-2-170280 JP-A-6-203202 JP 11-328408 A

  The present invention provides an image inclination detection device capable of detecting an image inclination with high accuracy by simple processing without requiring a large amount of processing time, and an image inclination for causing a computer to execute the function of such an image inclination detection device. The object is to provide a detection processing program.

  The present invention performs multi-layer wavelet space transformation on an image whose inclination is to be detected, and detects the angle of the image by performing Hough transform on at least a low frequency block (LL block) in order from the deepest layer. It is characterized by going.

  More specifically, a two-dimensional histogram of the ρ-θ plane obtained by the Hough transform is created, a detection angle that is the maximum frequency and a variance value of the detection angle are obtained, and when the variance value is equal to or less than a predetermined threshold value, The detected angle is detected as the angle of the image in the hierarchy. Further, when the variance value is larger than a predetermined threshold value, one of the blocks (LH block or HL block) in which either the vertical or horizontal direction is a high frequency and the other is a low frequency is detected from the detection angle having the maximum frequency. You can select and Hough transform to detect the angle in that hierarchy. Alternatively, blocks other than the low-frequency block can be synthesized, and the Hough transform can be performed on the candidate range including the detection angle having the maximum frequency, and the angle in the hierarchy can be detected. Such processing may be performed in order from a deep hierarchy to a shallow hierarchy.

  Further, the present invention performs wavelet space conversion on an image whose inclination is to be detected, detects a rough angle by performing a Hough transform on a low frequency block (LL block) in the wavelet space, and then vertically or horizontally from the rough angle. One of the blocks whose direction is high frequency and the other is low frequency (LH block or HL block) is selected and subjected to Hough transform to detect the angle of the image. Alternatively, after detecting the approximate angle in the same manner, blocks other than the low-frequency block (LH, HL, HH block) are synthesized, and the Hough transform is performed on the candidate range including the approximate angle, and the angle of the image is detected. It is what. Also in this case, it is possible to perform multi-level wavelet space conversion, perform processing in order from the deepest level, and detect more detailed angles.

  Furthermore, the present invention is characterized by causing a computer to execute the function of the image tilt detection apparatus as described above.

  According to the present invention, by performing wavelet space transformation on multiple hierarchies, the size of the low frequency block is reduced in the deep hierarchies, so that high-speed angle detection processing is possible. Further, for example, even when a halftone dot portion exists in the image, the halftone dot is crushed and noise is also crushed or removed, so that the angle can be detected without being affected by these. At this time, the angle can be detected without being influenced by noise even by using the low frequency block. In addition, since the block size is small in the deep hierarchy, the accuracy of the image angle cannot be obtained with only the low-frequency block in the deep hierarchy, but hierarchical angle detection that obtains angles in detail from the deep hierarchy in order. By performing the above, there is an effect that the inclination of the image can be obtained with high accuracy.

  In addition, after detecting the angle by Hough transforming the low frequency block, either one of the block (LH block or HL block) in which either the vertical or horizontal direction is high frequency and the other is low frequency is selected. Even if it is configured to detect the angle by Hough transform, or to detect a more detailed angle by synthesizing blocks other than the low frequency block, it is possible to obtain the inclination of the image with high speed and accuracy. effective.

  FIG. 1 is a block diagram showing an embodiment of the present invention. In the figure, 1 is a wavelet transform unit, and 2 is an angle detection unit. The wavelet transform unit 1 performs multi-layer wavelet space transform on an image whose inclination is to be detected.

  The angle detection unit 2 performs Hough transform on at least the low frequency block (LL block) of each layer converted by the wavelet transform unit 1 and detects the angle of the image from the conversion result. More specifically, the low-frequency block obtained as a result of wavelet space transformation is binarized and subjected to Hough transformation, and a two-dimensional histogram of the obtained ρ-θ plane is created. The variance value of each detection angle is obtained from this two-dimensional histogram. If the dispersion value is small, it indicates that the points are collected on the ρ-θ plane, and it can be determined that the angle is fixed to some extent. Therefore, when the variance value is equal to or less than a predetermined threshold, the detection angle that is the maximum frequency is detected as the image angle. When the variance value is larger than a predetermined threshold, it indicates a case where the angle cannot be specified, and in that case, a block other than the low frequency block can be used. For example, a maximum frequency angle is obtained from a two-dimensional histogram obtained by Hough transform of a low-frequency block, and a block other than the low-frequency block has a high frequency in the vertical or horizontal direction and the other is a low-frequency block ( LH block or HL block) is selected, binarized and Hough transformed, and an angle is detected within a predetermined range from the angle obtained from the low frequency block. Since the LH block contains a lot of horizontal components and the HL block contains a lot of vertical components, the LH block is used if the angle of the maximum frequency obtained from the low frequency block is close to the horizontal, and the HL block is used if the angle is close to the vertical. More detailed angle detection may be performed.

  Then, in such a process, the angles are determined hierarchically in order from the deepest hierarchy to the shallowest hierarchy. At this time, based on the angle obtained in a certain hierarchy, a candidate range including the angle is set, and in the next hierarchy, the angle is detected by performing a Hough transform at a fine angle on the candidate range. As a result, the detected angle is gradually corrected to the correct angle, and the tilt angle of the image can be finally detected with high accuracy.

  The above-described embodiment of the present invention will be further described. FIG. 2 is an explanatory diagram of wavelet space transformation performed in the wavelet transformation unit 1. Reference numerals 11 to 16 denote filter units. A general wavelet space transformation is performed by a configuration as shown in FIG. Here, only one layer is shown. The filter units 11 to 16 perform filter processing and subsampling. 'H' performs high-pass filter processing and 'G' performs low-pass filter processing. Further, “↓ 2” indicates that sub-sampling is performed to thin out one pixel for two pixels.

  First, for every two pixels adjacent to the input image in the main scanning direction, the filter unit 11 performs high-pass filter processing to thin out one pixel. The filter unit 12 performs low-pass filter processing to thin out one pixel.

  The image processed by the filter unit 11 is then subjected to high-pass filter processing by the filter unit 13 every two pixels adjacent in the sub-scanning direction, and one pixel is thinned out. As a result, high-pass filter processing is performed in both the main scanning direction and the sub-scanning direction, and an image (HH block) having a half resolution in both the main scanning direction and the sub-scanning direction is generated. Similarly, the filter unit 14 performs low-pass filter processing and thins out one pixel. As a result, an image (HL block) having a half resolution in both the main scanning direction and the sub-scanning direction, which has been subjected to high-pass filter processing in the main scanning direction and low-pass filter processing in the sub-scanning direction, is generated.

  Further, the image processed by the filter unit 12 is subjected to high-pass filter processing by the filter unit 15 every two pixels adjacent in the sub-scanning direction, and one pixel is thinned out. As a result, an image (LH block) having a half resolution in both the main scanning direction and the sub-scanning direction, which has been subjected to low-pass filter processing in the main scanning direction and high-pass filter processing in the sub-scanning direction, is generated. Similarly, the filter unit 16 performs low-pass filter processing and thins out one pixel. Thereby, low pass filter processing is performed in both the main scanning direction and the sub-scanning direction, and an image (LL block) having a half resolution in both the main scanning direction and the sub-scanning direction is generated.

  By such wavelet space transformation, four blocks of LL block, HL block, LH block, and HH block are generated from the image shown in FIG. 2B as shown in FIG. 2C. As can be seen from the generation process as described above, high frequency components are separated from the HH block, and image components that change finely are separated. Further, for example, noise and the like are also separated into the HH blocks. Since the HL block has been subjected to high-pass filter processing in the main scanning direction and low-pass filter processing in the sub-scanning direction, the main scanning direction changes finely, but components that change slowly in the sub-scanning direction are separated. . For example, a line segment extending in the sub scanning direction is separated into the HL blocks. On the other hand, the LH block has undergone low-pass filter processing in the main scanning direction and high-pass filter processing in the sub-scanning direction, so the change in the main scanning direction is gradual and components that change finely in the sub-scanning direction are separated. Is done. For example, a line segment extending in the main scanning direction is separated into the LH blocks. Since the LL block performs low-pass filter processing in both the main scanning direction and the sub-scanning direction, low frequency components are separated, and the tendency of the entire image is reflected in the LL block. Therefore, by referring to this LL block, it can be seen what the tendency of the entire image is, and in the present invention, the angle is detected from this LL block.

  When performing multi-layer wavelet space transformation, the processing may be repeated using the LL block of the four blocks obtained as described above as the input of the configuration shown in FIG. FIG. 2D shows the result of performing two-layer wavelet space transformation. The first result is indicated by “1”, and the second result is indicated by “2”. Wavelet space transformation is performed again on the LL1 block indicated by the bold line, and LL2, HL2, LH2, and HH2 blocks are generated. Of course, three or more wavelet space transformations may be performed in the same manner. For convenience of explanation, in the following example, two-layer wavelet space transformation is performed. The wavelet transform unit 1 performs the general multi-layer wavelet space transform as described above on the input image for tilt detection, and passes the result to the angle detection unit 2.

  The angle detection unit 2 receives the image subjected to wavelet space conversion passed from the wavelet conversion unit 1, detects angles in order from the lowest layer, and obtains more detailed angles. In the angle detection in each layer, the LL block from which at least low frequency components are separated is subjected to Hough transform, and the angle of the image is detected from the conversion result.

FIG. 3 is an explanatory diagram of a general Hough transform. The Hough transform converts a straight line in the xy coordinate system to one point in the ρ-θ coordinate system (polar coordinate system). The conversion formula is
ρ = X · cos θ + Y · sin θ
It is. If ρ and θ are uniquely determined (that is, one point in the ρ-θ coordinate system), the linear expression in X and Y becomes a straight line in the xy coordinate system.

  In the Hough transform, one point (X, Y) in the xy coordinate system becomes a trigonometric function expression that depends on X and Y from the above-described conversion formula, and becomes a curve in the ρ-θ coordinate system. At this time, as shown in FIG. 3A, the curves in the ρ-θ coordinate system for a plurality of points on the straight line in the xy coordinate system intersect at one point as shown in FIG. Therefore, if this intersection is found and a straight line in the xy coordinate system is obtained, the angle of the straight line component existing in the image can be obtained.

  Of course, since there are various linear components in the image, there are many intersections in the ρ-θ coordinate system. Therefore, for example, the trajectory of the curve converted for each black pixel of the binary image is counted for each coordinate in the ρ-θ coordinate system to create a two-dimensional histogram in the ρ-θ coordinate system, and the histogram peak, variance value, etc. From this, the linear component in the image is estimated and the angle is detected.

  Since the angle detection unit 2 uses the LL block as a processing target for the Hough transform, it is not easily affected by high frequency components, particularly noise, and the resolution is low. A peak can be easily formed in a histogram, and an angle can be detected as a tendency of the entire image. For example, halftone dots are separated into HH blocks as high-frequency components, and LL blocks are crushed or deleted during low-pass filter processing and subsampling, so that each halftone dot affects angle detection. There is no.

  Further, the size of the LL block is 1/4 of the original image (or the LL block in the next higher layer), and since the number of pixels to be converted is small, processing can be performed at high speed. For example, in the LL2 block shown in FIG. 2D, there are only 1/16 pixels of the original image, and processing can be performed at high speed.

  For example, when the wavelet transform unit 1 performs two-level wavelet space transform, the angle detection unit 2 can perform the Hough transform on the LL2 block as described above to detect the angle. However, the LL2 block is only 1/16 of the original image as described above, and the sub-sampling also causes collapse, so the accuracy of the obtained angle may be low. At this time, when the variance value of the peak is obtained in the above-described two-dimensional histogram, when the variance value is small (for example, when the variance value is equal to or smaller than a predetermined threshold value), it is shown that the peak appears clearly. It can be said that the obtained angle is accurately calculated. However, when the variance value is large (for example, when the variance value is larger than a predetermined threshold value), the peak is not so clearly displayed, and thus the obtained angle may not be obtained with high accuracy. In such a case, processing for obtaining the angle with higher accuracy is performed.

  Furthermore, as a method for obtaining the angle with high accuracy, it is possible to increase the accuracy by using the LL block in the upper layer and performing Hough transform in the candidate range including the angle already obtained. Of course, the process may be shifted to the process of the next higher level as it is, but here, the process of increasing the accuracy is performed using another block of the same level before moving to the process of the next higher level.

  As described with reference to FIG. 2, when wavelet space conversion is performed, the HL block includes a vertical component and the LH block includes a horizontal component. Therefore, if the angle of the image corresponding to the point of the maximum frequency obtained from the two-dimensional histogram on the ρ-θ plane is close to the vertical, the HL block is used, and if the angle is close to the horizontal, the LH block is used. Can be detected. For example, if the angle obtained from the LL block is 0 ° to 45 ° or 135 ° to 180 °, it is judged to be almost horizontal and the LH block is used, and if it is 45 ° to 135 °, it is judged to be nearly vertical. An angle may be detected by performing Hough transform using an HL block. At this time, Hough transform and angle detection may be performed in a candidate range including the angle obtained from the LL block.

  Alternatively, the HL block, the LH block, and the HH block excluding the LL block can be combined, and the Hough transform and the angle detection can be performed within the candidate range including the angle obtained from the LL block. For example, when the image is crushed or white spots are largely generated by low-pass filter processing and sub-sampling, it is possible to detect a more detailed angle by using a high-frequency component. At this time, if a candidate range is set, a completely different angle is not detected due to noise or the like.

  As described above, an angle obtained using a block other than the LL block is set as an angle in the hierarchy, and the angle of the image can be obtained with high accuracy by correcting the angle in the hierarchy one level higher. For example, when the variance value of the peak obtained from the above-described two-dimensional histogram is sufficiently small, the final image angle can be output without performing processing for higher layers. If sufficient accuracy is obtained, the angle of the final image may be obtained by only angle detection from the LL block and angle detection from blocks other than the LL block in only one layer.

  4 and 5 are flowcharts showing an example of the operation in the embodiment of the present invention. Here, as shown in FIG. 2D, an example of processing performed by the angle detection unit 2 when two-layer wavelet space conversion is performed by the wavelet conversion unit 1 is shown. 4 shows the angle detection process for the lowermost layer (LL2, HL2, LH2, HH2), and FIG. 5 shows the angle detection process for the upper layer (LL1, HL1, LH1, HH1).

  The angle detection unit 2 first performs angle detection processing using the LL2 block, which is the lowest frequency block at the lowest level. In S51, the LL2 block is binarized. In S52, the Hough transform is performed on the binarized LL2 block to create a two-dimensional histogram on the ρ-θ plane. The two-dimensional histogram created at this time needs to be created in the range of 0 ° ≦ θ ≦ 180 °. Since the range is wide, it is only necessary to determine the degree of accuracy with which the angle is obtained in consideration of the processing speed and the like. For example, a histogram of several degrees may be used. Here, a two-dimensional histogram was created in units of 3 °.

In S53, the maximum frequency (ρ, θ) is obtained from the two-dimensional histogram created in S52. This is _defined as max (ρ _LL2 , θ _LL2 ). Also, a variance value V _LL2 is calculated for this max (ρ _LL2 , θ _LL2 ). The dispersion value V _LL2 may be obtained only for θ _LL2 .

In S54, it is determined whether or not the variance value V _LL2 for max (ρ _LL2 , θ _LL2 ) obtained in S53 is larger than a predetermined threshold value TH. as can be detected in _{S55, max (ρ LL2, θ} LL2) is output as the inclination angle of the input image to theta _LL2 of, the process ends. As a result, the subsequent processing does not have to be performed, and the processing time can be shortened. In addition, in order to obtain | require an angle in more detail, you may transfer to the process (S71) of LL1 block.

If the variance value V _LL2 is larger than the predetermined threshold TH, it is assumed that θ _LL2 is not accurate enough to adopt the angle of the image, and in this example, blocks other than the LL block in the same hierarchy, particularly the HL block or LH Using any of the blocks, the angle is further determined. In _{S56, max (ρ LL2, θ} LL2) of theta _LL2 is 0 ° <θ _LL2 <45 ° or 135 ° <θ _LL2 <180 ° determines whether the. In the case of this angle, since the inclination is almost horizontal, the processing target is set as the LH2 block in S57, and the LH2 block is binarized. In S58, a Hough transform is performed on the binarized LH2 block to create a two-dimensional histogram on the ρ-θ plane. The two-dimensional histogram created at this time may be created with the same accuracy as in S52. Alternatively, a two-dimensional histogram may be created with increased accuracy within the range of the condition of S56. Here, as in S52, a two-dimensional histogram was created in units of 3 °.

In S59, the maximum frequency (ρ, θ) is obtained from the two-dimensional histogram created in S58. Let this be max (ρ _LH2 , θ _LH2 ). Also, a variance value V _LH2 is calculated for this max (ρ _LH2 , θ _LH2 ). Note that the dispersion value V _LH2 may be obtained for θ _LH2 .

In S60, it is determined whether or not the variance value V _LH2 for max (ρ _LH2 , θ _LH2 ) obtained in S59 is greater than a predetermined threshold value TH. as can be detected in _{S61, max (ρ LH2, θ} LH2) outputs as the tilt angle of the input image to theta _LH2 of, the process ends. Also in this case, the processing time for inclination detection can be shortened by ending with the processing so far. In addition, in order to obtain | require an angle in more detail, you may transfer to the process (S71) of LL1 block.

If the variance value V _LH2 is larger than the predetermined threshold value TH, it is determined that the angle is not sufficiently accurate to adopt θ _LH2 as the image angle, and the process proceeds to the processing of the LL1 block (S71) in order to obtain the angle in more detail. To do.

S56 in max (ρ _{LL2, θ LL2)} if the theta _LL2 is determined not to be 0 ° <θ _LL2 <45 ° or _{135 ° <θ LL2 <180 °} , i.e. when the 45 ° <θ _LL2 <135 ° Since the inclination is nearly vertical, the processing target is set as the HL2 block in S62, and the HL2 block is binarized. In S63, a Hough transform is performed on the binarized HL2 block to create a two-dimensional histogram on the ρ-θ plane. The two-dimensional histogram created at this time may be created with the same accuracy as in S58.

In S64, the maximum frequency (ρ, θ) is obtained from the two-dimensional histogram created in S63. Let this be max (ρ _HL2 , θ _HL2 ). Also, a variance value V _HL2 is calculated for this max (ρ _HL2 , θ _HL2 ). Note that the variance value V _HL2 may be obtained for θ _HL2 .

In S65, it is determined whether or not the variance value V _HL2 for max (ρ _HL2 , θ _HL2 ) obtained in S64 is greater than a predetermined threshold value TH. If the variance value is equal to or less than the threshold value TH, the angle is sufficiently accurate. as can be detected in _{S66, max (ρ HL2, θ} HL2) is output as the inclination angle of the input image to theta _HL2 of, the process ends. Thereby, the detection processing time of the inclination angle can be shortened. In addition, in order to obtain | require an angle in more detail, you may transfer to the process (S71) of LL1 block.

If the variance value V _HL2 is larger than the predetermined threshold value TH, it is determined that the angle is not accurate enough to adopt θ _HL2 as the image angle, and the process proceeds to the processing of the LL1 block (S71) in order to obtain the angle in more detail. To do.

  Next, an angle detection process for the next hierarchy (LL1, HL1, LH1, HH1) is performed based on the angles obtained so far by the processes after S71. In S71, the LL1 block is binarized. In S72, the Hough transform is performed on the binarized LL1 block to create a two-dimensional histogram on the ρ-θ plane. The two-dimensional histogram created at this time creates a two-dimensional histogram in detail for the candidate range including the angle obtained in S79 and S84 (or S73). For example, a range of ± 3 ° centered on the previously obtained angle is set as a candidate range, and a two-dimensional histogram is created in units of 0.1 °.

In S73, the maximum frequency (ρ, θ) is obtained from the two-dimensional histogram created in S72. This is _defined as max (ρ _LL1 , θ _LL1 ). Further, a variance value V _LL1 for this max (ρ _LL1 , θ _LL1 ) is calculated. The variance value V _LL1 may be obtained only for θ _LL1 .

In S74, it is determined whether or not the variance value V _LL1 for max (ρ _LL1 , θ _LL1 ) obtained in S73 is larger than a predetermined threshold value TH. as can be detected in S75, and outputs a tilt angle of max (ρ _{LL1, θ LL1)} entered the theta _LL1 of the image, the process ends. If the tilt angle can be output at this time, the processing time can be shortened by the amount that the subsequent processing is not performed.

If the variance value V _LL1 is larger than the predetermined threshold value TH, it is assumed that the accuracy is not sufficient to adopt θ _LL1 as the angle of the image. In this example, blocks other than the LL block of the same hierarchy, particularly the HL block or LH Using any of the blocks, the angle is further determined. In _{S76, max (ρ LL1, θ} LL1) θ LL1 of determining whether 0 ° <θ _LL1 <either 45 ° or _{135 ° <θ LL1 <180 °} . In the case of this angle, since the inclination is almost horizontal, the processing target is set to the LH1 block in S77, and the LH1 block is binarized. In S78, the Hough transform is performed on the binarized LH1 block to create a two-dimensional histogram on the ρ-θ plane. The two-dimensional histogram created at this time may be created with the same accuracy as in S72. Alternatively, the candidate range may be narrowed based on the angle obtained in S73, and the two-dimensional histogram may be created with improved accuracy. Here, a two-dimensional histogram is created in units of 0.1 ° in the same candidate range as S72.

In S79, the maximum frequency (ρ, θ) is obtained from the two-dimensional histogram created in S78. Let this be max (ρ _LH1 , θ _LH1 ). In S80, max (ρ _LH1 , θ _LH1 ) θ _LH1 is output as the tilt angle of the input image, and the process ends.

S76 in max (ρ _{LL1, θ LL1)} If theta _LL1 of is determined not to be 0 ° <θ _LL1 <45 ° or _{135 ° <θ LL1 <180 °} , i.e. when the 45 ° <θ _LL1 <135 ° Since the inclination is nearly vertical, the processing target is set as the HL1 block in S81, and the HL1 block is binarized. In S82, a Hough transform is performed on the binarized HL1 block to create a two-dimensional histogram on the ρ-θ plane. The two-dimensional histogram created at this time may be created with the same candidate range and accuracy as in S78.

In S83, the maximum frequency (ρ, θ) is obtained from the two-dimensional histogram created in S82. Let this be max (ρ _HL1 , θ _HL1 ). In _{S84, max (ρ HL1, θ} HL1) is output as the inclination angle of the input image to theta _HL1 of, the process ends.

  As described above, the angle of the image can be detected with high accuracy by performing multi-layer wavelet transform and detecting the angle hierarchically. In addition, the angle can be detected at a higher speed than when the angle is detected with the same accuracy as the original image.

  In the above operation example, threshold values for determining a variance value are not distinguished, but they may be the same, or different threshold values may be used for each layer or LL block of each layer and LH and HL blocks. May be. Of course, each can be set arbitrarily.

  In this example, the case of two layers is shown, but the angle can be detected hierarchically in the same manner even if there are three or more layers. Furthermore, it is possible to detect angles hierarchically only by the LL block without using the LH block or the HL block. Alternatively, the LH block, the HL block, and the HH block may be combined and Hough transformed, a two-dimensional histogram may be created, and the angle of the layer may be obtained from the maximum frequency (ρ, θ).

  FIG. 6 is an explanatory diagram of an example of a computer program and a storage medium storing the computer program when the function of the image tilt detection apparatus of the present invention is realized by a computer program. In the figure, 21 is a program, 22 is a computer, 31 is a magneto-optical disk, 32 is an optical disk, 33 is a magnetic disk, 34 is a memory, 41 is a magneto-optical disk apparatus, 42 is an optical disk apparatus, and 43 is a magnetic disk apparatus.

  Part or all of the functions of the units described in the above embodiments can be realized by a program 21 that can be executed by a computer. In this case, the program 21 and data used by the program can be stored in a computer-readable storage medium. A storage medium is a signal format that causes a state of change in energy such as magnetism, light, electricity, etc. according to the description of a program to a reader provided in the hardware resources of a computer. Thus, the description content of the program can be transmitted to the reading device. For example, there are a magneto-optical disk 31, an optical disk 32 (including a CD and a DVD), a magnetic disk 33, a memory 34 (including an IC card, a memory card, and the like). Of course, these storage media are not limited to portable types.

  By storing the program 21 in these storage media and mounting these storage media in, for example, the magneto-optical disk device 41, the optical disk device 42, the magnetic disk device 43, or a memory slot (not shown) of the computer 22, The program 21 can be read to execute the function of the image tilt detection apparatus of the present invention. Alternatively, the program 21 may be transferred to the computer 22 via a network or the like and executed.

  Of course, some functions may be configured by hardware, or all may be configured by hardware. Alternatively, it can be configured as a program including the present invention together with other configurations. For example, it may be configured as one program together with a control program in an image forming apparatus such as a copying machine so as to detect the inclination of the document. Of course, in the case of application to other purposes, integration with a program for that purpose is also possible.

It is a block diagram which shows one Embodiment of this invention. It is explanatory drawing of the wavelet space transformation performed in the wavelet transformation part 1. FIG. It is explanatory drawing of general Hough conversion. It is a flowchart which shows an example of the operation | movement in one Embodiment of this invention. It is a flowchart (continuation) which shows an example of the operation | movement in one Embodiment of this invention. It is explanatory drawing of an example of the storage medium which stored the computer program in the case of implement | achieving the function of the image inclination detection apparatus of this invention with a computer program, and the computer program.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Wavelet transformation part, 2 ... Angle detection part, 11-16 ... Filter part, 21 ... Program, 22 ... Computer, 31 ... Magneto-optical disk, 32 ... Optical disk, 33 ... Magnetic disk, 34 ... Memory, 41 ... Magneto-optical Disk device 42... Optical disk device 43.

Claims (9)

  An image comprising: transform means for performing multi-layer wavelet space transform on an image whose inclination is to be detected; and detection means for detecting an angle of the image by performing Hough transform on at least a low-frequency block in each layer. Tilt detection device.   The detecting means detects the angle by performing a Hough transform on the deepest low-frequency block and then corrects the angle by performing a Hough transform on the low-frequency block from the deeper side in order to a predetermined accuracy. The image inclination detection apparatus according to claim 1.   The detection means performs a Hough transform at a finer angle as the hierarchy is shallower, sets a candidate range based on an angle obtained at a previous hierarchy for a certain hierarchy, and performs a Hough transformation at a fine angle for the candidate range. The image inclination detection device according to claim 2, wherein the angle is obtained by performing the calculation.   The detection means creates a two-dimensional histogram of the ρ-θ plane obtained by the Hough transform, obtains a dispersion value of each detection angle, and determines a detection angle that has a maximum frequency when the dispersion value is equal to or less than a predetermined threshold value. The image inclination detection device according to any one of claims 1 to 3, wherein the image inclination detection device detects the angle of the image.   The detection means obtains a detection angle having a maximum frequency and a variance value of the detection angle from a two-dimensional histogram of the ρ-θ plane obtained by the Hough transform for the low frequency block, and when the variance value is larger than a predetermined threshold value. Is characterized in that either one of a block (LH block or HL block) having either a high frequency in the vertical or horizontal direction and a low frequency in the other is selected from the detected angle and subjected to Hough transform to detect the angle. The image inclination detection apparatus according to any one of claims 1 to 4.   The detection means obtains the detection frequency and the variance value of each detection angle from the two-dimensional histogram of the ρ-θ plane obtained by the Hough transform for the low frequency block, and when the variance value is larger than a predetermined threshold value 5. The image inclination according to claim 1, wherein blocks other than the low-frequency block are synthesized and Hough transform is performed on a candidate range including the detection angle, and the angle is detected. Detection device.   A conversion means for performing wavelet space conversion on an image whose inclination is to be detected, and a Hough transform of a low-frequency block in the wavelet space to detect an approximate angle, and either the vertical or horizontal direction from the approximate angle has a high frequency and the other An image inclination detection apparatus comprising: a detecting unit that selects any one of low-frequency blocks (LH block or HL block), performs Hough transform, and detects the angle of the image.   A conversion unit that performs wavelet space conversion on an image whose inclination is to be detected, and a candidate that includes the approximate angle by synthesizing other blocks after detecting an approximate angle by Hough transforming a low-frequency block in the wavelet space An image tilt detection apparatus comprising detection means for performing a Hough transform on a range and detecting an angle of the image.   An image inclination detection processing program for causing a computer to execute a process of detecting an inclination of an input image, wherein the function of the image inclination detection device according to any one of claims 1 to 8 is executed on a computer. An image inclination detection processing program characterized in that

Images