JP6863753B2 - Devices, methods and computer-readable storage media that determine the angle of rotation of text - Google Patents

Description

The present disclosure generally relates to areas of text orientation.

Typically, the estimation of the angle at which the character is drawn is made using an optical character recognition (OCR) process. This process is slow and unsuitable for real-time OCR applications. The existing solution is to ignore such a solution or require the user to manually specify the angle, and manually specifying the angle is especially of the text that needs to be processed. It is troublesome for the user when the amount is large.

Devices, methods and computer-readable storage media that determine the angle of rotation of text. The method is to calculate the distance to the closest object, to calculate the average distance to the closest object, and to average for each object of multiple objects contained in the text in the image. Determining the ratio between the distance and the average font line width, which is the average of the font line widths of each of multiple objects, and determining the angle of rotation of the text by comparing the ratio with the threshold. To be equipped with.

The disclosure will be better understood by reading the description below and looking at the accompanying drawings. These drawings are provided only as an unlimited example of embodiments.

Four different angles of rotation according to one embodiment are shown. An example of portrait text according to one embodiment is shown. A flowchart of a process for determining whether a text block according to an embodiment is portrait or landscape is shown. A flowchart of a process for determining the rotation angle of (one or more) vertical blocks according to an embodiment is shown. An object according to an embodiment is shown. A text block according to an embodiment is shown. An example of a process executed by one embodiment is shown. A computer system capable of realizing the (one or more) embodiments of the present disclosure is shown.

One embodiment is a device that determines the angle of rotation of text in an image, comprising a circuit, the circuit providing, for each object of a plurality of objects contained in the text, the distance to the closest object. Calculate, calculate the average distance to the closest object, determine the ratio between the average distance and the average font line width, which is the average of the font line widths of each of multiple objects, and the ratio and threshold. Note the device configured to determine the angle of rotation of the text by comparing with.

In one embodiment, the circuit is further configured to determine the circumscribed rectangle of each object, the circumscribed rectangle surrounds each object and touches the outermost of each object, and the circuit is closest to each circumscribed rectangle. Calculate the distance to the closest object by calculating the distance to the circumscribed rectangle.

In one embodiment, the circuit determines, for each object, an object area that corresponds to the entire print area of each object in the image, and for each object, determines an object outer circumference that corresponds to the outer circumference of each object. It is further configured to calculate the font line width of each object by dividing the object area by the outer circumference of the object and multiplying by 2.

In one embodiment, the circuit is configured to calculate the average font line width by dividing the sum of the font line widths of the plurality of objects by the total number of the plurality of objects.

In one embodiment, the circuit determines one or more text blocks of text in an image before calculating the distance to the closest object to each object, where each text block is a plurality of objects. For each text block, determine whether the text block is portrait or landscape, and in response to the determination that the text block is portrait, for each object in the text block. It is configured to calculate the distance to the closest object.

One embodiment is to calculate the distance to the closest object and the average distance to the closest object for each object of a plurality of objects contained in the text in the image. The angle of rotation of the text is determined by determining the ratio between the average distance and the average font line width, which is the average of the font line widths of each of a plurality of objects, and by comparing the ratio and the threshold value. Pay attention to what to do and how to prepare for it.

In one embodiment, the method is to determine the circumscribing rectangle of each object, which further comprises surrounding each object and contacting the outermost of each object, with respect to the closest object. The calculation of the distance is done by calculating the distance between each circumscribing rectangle and the closest object.

In one embodiment, the method is to determine, for each object, an object area that corresponds to the entire print area of each object in the image, and for each object, an object outer circumference that corresponds to the outer circumference of each object. And to calculate the font line width of each object by dividing the object area by the outer circumference of the object and multiplying by 2.

In one embodiment, the method further comprises calculating the average font line width by dividing the sum of the font line widths of the plurality of objects by the total number of the plurality of objects.

In one embodiment, the method is to determine one or more text blocks of text in an image before calculating the distance to the closest object to each object, where each text block is , In response to the decision to contain multiple objects, to determine for each text block whether the text block is portrait or landscape, and to determine that the text block is portrait, the text block It further comprises calculating the distance to the closest object for each object in.

In one embodiment, the method further comprises performing optical character recognition (OCR) after determining the angle of rotation of the text.

One embodiment is to calculate the distance to the closest object for each object of a plurality of objects contained in the text in the image when executed on a computer, and to the closest object. Calculating the average distance of distances, determining the ratio between the average distance and the average font line width, which is the average of the font line widths of each of multiple objects, and comparing the ratio with the threshold. Focus on a non-temporary computer-readable storage medium that has computer-executable instructions that cause a computer to perform a method of determining the rotation angle of a text by means of.

In one embodiment, the non-temporary computer-readable storage medium is to determine the circumscribing rectangle of each object, which further comprises surrounding each object and contacting the outermost of each object. Calculating the distance to the closest object is done by calculating the distance between each circumscribing rectangle and the closest object.

In one embodiment, the non-temporary computer-readable storage medium determines, for each object, an object area that corresponds to the entire print area of each object in the image, and for each object, each object. It further comprises determining the outer circumference of the object corresponding to the outer circumference of the object, and calculating the font line width of each object by dividing the object area by the outer circumference of the object and multiplying by 2.

In one embodiment, the non-temporary computer-readable storage medium further comprises calculating the average font line width by dividing the sum of the font line widths of the plurality of objects by the total number of the plurality of objects.

In one embodiment, the non-temporary computer-readable storage medium is to determine one or more text blocks of text in an image before calculating the distance to the closest object to each object. Each text block contains multiple objects, determines for each text block whether the text block is portrait or landscape, and determines that the text block is portrait. In response, it further comprises calculating the distance to the closest object for each object in the text block.

In one embodiment, the non-temporary computer-readable storage medium further comprises performing optical character recognition (OCR) after determining the angle of rotation of the text.

The present disclosure focuses on a method of rapidly estimating or determining the angle at which a character (eg, an ideographic character) is written in an image of a vertically flowing text block. The angle written in this way is called the angle of rotation. It should be noted that rotation and orientation may be used interchangeably throughout this disclosure.

As shown in FIG. 1, in one embodiment, there are four different angles of rotation. FIG. 1A shows characters with a rotation angle of 0 °. FIG. 1B shows characters with a rotation angle of 90 °. FIG. 1C shows characters with a rotation angle of 180 °. FIG. 1D shows characters with a rotation angle of 270 °.

Text that flows vertically instead of horizontally is called portrait text. FIG. 2 shows four examples of portrait text. FIG. 2A shows an example in which the text of a language derived from Latin is rotated by 90 °. FIG. 2B shows an example in which the text of a language derived from Latin is rotated by 270 °. FIG. 2C shows an example of text in a language derived from Latin written vertically without rotating the letters (rotated by 0 °), and FIG. 2D is written vertically without rotating the letters. An example of Japanese text is shown.

Consider the text to be landscape in relation to the horizontal axis of the image (from left to right in the image) and the text to be vertical in relation to the vertical axis of the image (from top to bottom of the image) Please note that The orientation of the landscape text and the orientation of the portrait text are shown in FIGS. 1 and 2. Vertical text is categorized in columns (top to bottom) of the image parallel to the vertical axis of the image, whereas landscape text is rows (left to right) of the image parallel to the horizontal axis of the image. It is classified in.

It is noted that the present disclosure makes decisions based on the size of the spacing between letters and the characteristics of the letters written.

There is a certain inconvenience in performing OCR for each character to determine rotation. For example, this process is slow and unsuitable for real-time OCR applications. As mentioned above, the existing solution is to ignore the angle of rotation of such text or require the user to manually specify the angle of rotation.

As described above, this embodiment describes a method of quickly and automatically estimating the rotation angle of portrait text. The embodiments of the present disclosure can be used to facilitate the OCR process, and other applications improve their accuracy in real time. For example, first the embodiments described herein may be used, after which the OCR may apply the results to properly recognize (one or more) characters. In other words, the angle of rotation of the text may first be determined as described later with reference to FIGS. 3 and 4, and then OCR may be applied to the text.

FIG. 3 shows a process of determining whether a text block according to one embodiment is portrait or landscape. Note that a given image (eg, an image of a scanned document) has, for example, vertically drawn (written) text and horizontally drawn (written) text. Such images may represent advertising surfaces, newspaper surfaces, leaflet surfaces, etc. or other types of pages / documents.

Therefore, in one embodiment, first, in step 100, an object (for example, a character or letter) in the image is detected. It should be noted that objects and letters may be used interchangeably in the present disclosure. Next, in step 110, the object is divided into blocks. The block may be, for example, any group of characters (eg, sentences, paragraphs, etc.). Blocks may be distinguished from other blocks based on whether the distance between blocks is greater than a predetermined distance. For example, the distance between two paragraphs is significantly greater than the distance between two letters in a word. Therefore, such distances are used in determining the size of the block. Note that in one embodiment, each block may be of a different size. Note that in one embodiment, some blocks (one or more) may be the same size while other blocks (one or more) may be different sizes (one or more).

Note that in one embodiment, the minimum distance between blocks may be twice the size of the characters and the distance between the characters may be 0.2 times or less the size of the characters. Note that these distances are pixels.

Next, in step 120, the orientation angle (ie, 0 °, 90 °, 180 ° or 270 °) is determined for each block. In one embodiment, the angles were discussed in a method using Springer-Verlag's "Combined orientation and skew detection-using geometry modeling" by Beusekom et al., Published January 16, 2010. It may be decided and the entire contents are incorporated here by reference. In one embodiment, the angle is determined by the "Fast seamless skew and orientation" method from the 2010 International Conference on Document Recognition published in January 2010 by IEEE, etc. The entire contents may be incorporated here by reference. In one embodiment, the angle may be determined by other methods (one or more) known to those of skill in the art. For example, any method that can detect the angle and the skew angle of the image may be used.

For the sake of brevity, the method discussed in Beusekom et al. (Hereinafter referred to as Beusekom) uses geometric matching to extract lines of text from a scanned document. A quality function is specified that gives the quality of alignment of the text line model for a given point. The purpose is to maximize the number of bounding boxes that match the model and to minimize the distance of each reference point from the reference value in the robust least squares squares parameter of each text line of the document image. Is to find a set of. A very important idea of Beusekom's method is to use ascender modeling as well as modeling descenders.

Model the x-line (a line that passes over non-ascender lowercase letters such as x, a, c, etc.) as a straight line, and place the ascender line at a predetermined distance above the x-line. However, it is modeled as a line parallel to the x line. Reference points {y1, y2 ,., Obtained by taking the middle of the top line of the boundary box of the connection part of the document image. .. .. , Yn}. The purpose of detecting lines of text is to find the maximum set width. In Latin-derived languages, script ascenders are more likely to occur than descenders, so components are more likely to align with ascender lines than descender lines. Descender / Ascender-consistent components receive fewer scores than reference / x-line-consistent components. Therefore, in general, the overall quality of the descender line is higher than the overall quality of the ascender line. This information is used to find the vertical direction of the page.

In the Beusekom method, the descender model is used to calculate the overall quality of n best lines, and then the ascender model is used to calculate the overall quality of n best lines. If the quality of the ascender model is higher than the quality of the descender model, the page is reported upside down (180 ° rotation). Calculating the ascender model for a given page image in the correct orientation is equivalent to calculating the descender model for a page rotated 180 °. Therefore, for any image, only the descender model of the original image and the descender model of the 180 ° rotated image are calculated. Images with even better descender quality are reported as images with the correct orientation.

This concept is easily extended to detected 90 ° orientation pages and 270 ° orientation pages. The landscape textline model does not fit well into portrait textlines, so the overall quality of the n optimal lines in portrait orientation for the correct portrait format (right side up portrait page) is It is significantly lower than the overall quality of n horizontal optimum lines. Therefore, the exact orientation of the page can be determined by calculating the descender quality by rotating the page in all four orientations.

Returning to FIG. 3, in step 130, each block is classified into either portrait orientation or landscape orientation based on the angle result. In particular, when the orientation of the block is 90 ° or 270 °, the block is classified as being vertically oriented. Otherwise (ie, when the orientation of the block is 0 ° or 180 °), the block is classified as sideways. When the block is oriented sideways, no further processing is required because the angle of rotation is the angle described above (ie, for example, the angle determined using the method discussed in Beusekom).

In step 140, it is determined whether or not all the blocks have been classified. If there are blocks that have not yet been classified, the process returns to step 120 and repeats until all blocks are classified.

If the block is classified as vertical, then the rotation angle is 0 ° (as shown in FIG. 2C), the rotation angle is 90 ° (as shown in FIG. 2A), or (as shown in FIG. 2B). The rotation angle of 270 ° is determined with reference to FIG. Therefore, in step 200, the font line width is calculated for each character in the text block using the following equation 1.

FIG. 5 shows an object area 300 and an object outer circumference 310 of the letter “L” (object 320) according to one embodiment. Note that the object area 300 means the print area of the character 320 (ie, the black portion of the character if the character is drawn in black). The object outer circumference 310 means the outer circumference of the character 320 (that is, the outline of the character / print area). In other words, the object area 300 corresponds to the number of foreground pixels, and the object outer circumference 310 corresponds to the outer circumference contour. As an example, the maximum font line width is 7 pixels with respect to the font size of Calibiri 11.

Next, in step 210, the average font line width of all the characters 320 of the block is calculated using the following equation 2.

Then, in step 220, for each object 320, a circumscribed rectangle 330 (that is, a box surrounding the character 320 and tangent to the outermost character) is calculated. The circumscribed rectangle 330 of the object “L” 320 is shown in FIG. Note that in one embodiment, the circumscribed rectangle 330 does not touch the character 320 but may surround the character 320 and be at a predetermined distance (eg, one pixel) from the character 320.

Further, in step 230, for each object 320, the dimension of the interval 340 with respect to the nearest object 320 is calculated (see FIG. 6). Note that in one embodiment, this dimension is calculated from the circumscribed rectangle 330 of one object 320 to the circumscribed rectangle 330 of another (adjacent) object 320. As shown in FIG. 6, the above-mentioned interval 340 represents the distance of the object 320 to the nearest adjacent object 320. Note that FIG. 6 shows an embodiment of a block of text in which objects 320 (ie, each of "L", "E", "A" and "D") have an interval of 340.

Next, in step 240, the average of the dimensions of all object spacings 340 is calculated. In step 250, the ratio between the average spacing dimension and the average font line width is determined using Equation 3 below.

Next, in step 260, the ratio is compared to the threshold T to determine the angle of rotation. In one embodiment, the threshold T may be a predetermined value corresponding to the font weight of most characters in a particular language. For example, in one embodiment, T may be set to a value of 3 for a language derived from Latin. The reason is that the characters in Latin-derived languages (such as English) are drawn in bold font. Note that the value 3 represents the number of pixels that make up the font weight. In one embodiment, T may be set to a value of 5 for Asian languages. The reason is that Asian languages (such as Japanese) are drawn in thin fonts (to write in more detail).

The threshold T described above is applicable to all styles of text (including bold and / or italicized text). Further, in one embodiment, the threshold may be changed based on the particular application applied. For example, a training framework such as Bayesian may be used to estimate threshold values that are suitable for a particular application. Such estimation may be performed by a system such as the system shown in FIG.

Therefore, if the determined ratio is T or greater (ie, ratio ≥ T), the text block is determined to be portrait orientation containing characters with a rotation angle of 0 ° or 180 degrees. Otherwise (ie, if the determined ratio is less than T (ratio <T)), the text block has an angle of rotation equal to the estimated angle of rotation, as determined in step 120 of FIG. Is considered to be vertically oriented.

In order to briefly summarize one embodiment according to the present disclosure, it is assumed that a sheet / page containing the characters shown in FIG. 7 is scanned using a scanner or a similar device. In one embodiment, first, an object (eg, one or more characters) in the scanned image is detected. Once these characters are detected, they are split into blocks. For example, the process determines the five blocks 400-440 of FIG.

Next, it is determined whether each block 400 to 440 is in landscape orientation or portrait orientation. Note that this decision is made for each block but not for each object (character) within each block. When performing the process shown in FIG. 3, blocks 400 and 410 are determined to be landscape (ie, landscape text) with respect to the image (scanned page), and blocks 420, 430 and 440 are images (scanned). It is determined to be portrait (ie, portrait text) with respect to the page.

Next, the process shown in FIG. 4 is applied to the block determined to be vertically oriented. In this process, as described above, each object (character) of blocks 420, 430 and 440 is analyzed. By performing the process described in FIG. 4, the block 420 has a rotation angle of 90 °, and the block 430 has a rotation angle of 0 ° (that is, the text is drawn vertically but the characters in the block are drawn. Is not rotating) and it is determined that the block 440 has a rotation angle of 270 °.

The angle of rotation (one or more) may then be used by the OCR process to improve accuracy and reduce the calculation time of the characters to be recognized. By having knowledge of the angle of rotation of a particular block of text, the OCR process knows a particular layout of the text (eg, one of the layouts shown in Figure 2) and there is no rotation when the character actually rotates. Do not waste calculation time and power when trying to recognize the assumed characters. As a result, the calculation time is shortened, power is saved, and the accuracy is improved, so that the performance of the device / machine that executes the above-described processing is improved.

In addition, embodiments of the present disclosure may be used to assist OCR in properly recognizing characters and / or to rotate the characters, for example, when a sheet containing text is improperly scanned. Please note that.

The embodiments of the present disclosure provide many differences from the prior art. For example, character recognition (eg, OCR) is not used to find the angle of rotation. Further, the angle of rotation is determined for each text block and not for the entire scanned image. In other words, according to one embodiment, the process does not easily determine that the entire image is also landscape because the block 400 of FIG. 7 is landscape. On the contrary, each block is examined individually to determine the exact orientation of each block. If the text should be rotated, rotate each block of text individually based on its orientation. Therefore, in one embodiment, the entire scanned image is not rotated based on the orientation of the detected blocks.

In addition, the embodiments of the present disclosure may be applied to languages derived from Latin, Asian languages, and the like. Further, the embodiments of the present disclosure may be applied to text of any font and size.

Each of the functions / methods / processes described in the embodiments may be realized by one or more processing circuits (or circuits). For example, the processes shown in FIGS. 3 and 4 may be realized by one or more processing circuits (or circuits). The processing circuit includes a programmed processor (eg, processor 1203 in FIG. 8), such as a processor that includes the circuit. Processing circuits also include devices such as application specific integrated circuits (ASICs) and conventional circuit components arranged to perform the listed functions.

FIG. 8 shows a computer system 1201 capable of implementing (one or more) embodiments of the present disclosure. The computer system 1201 may be a general-purpose computer or a specific dedicated machine. In one embodiment, computer system 1201 becomes a particular dedicated machine when programming processor 1203 to determine the angle of rotation of an object / text. The computer system 1201 may be implemented as (or as part of) a multifunction peripheral device (MFP), an integrated printer, a scanner, or the like.

The computer system 1201 includes a magnetic hard disk 1207 and a removable media drive 1208 (for example, a floppy disk (registered trademark) drive, a read-only compact disk drive, a read / write compact disk drive, a compact disk jukebox, a tape drive, and a removable optical magnetic drive). Includes a disk controller 1206 coupled to bus 1202 to control one or more storage devices that store information and instructions such as. The storage device should have a suitable device interface (eg, Small Computer Interface System (SCSI), integrated device electronics (IDE), Enhanced IDE (E-IDE), Direct Memory Access (DMA) or Ultra DMA). It may be added to the computer system 1201 by using it.

The computer system 1201 may also include a dedicated logic device (eg, an ASIC) or a configurable logic device (eg, a simple programmable logic device (SPLD), a complex programmable logic device (CPLD) and a field programmable gate array (FPGA)). ..

The computer system 1201 may also include a display controller 1209 coupled to a bus 1202 to control a display 1210, which can be a liquid crystal display (LCD) displaying information to a computer user. Note that the display 1210 may be a touch panel display. The computer system includes input devices such as a keyboard 1211 and a pointing device 1212 that exchange information with the computer user and provide the information to the processor 1203. The pointing device 1212 may be, for example, a finger on a mouse, trackball, touch screen sensor, or a pointing stick that transmits instruction information and command selection to processor 1203 and controls cursor movement on display 1210.

Computer system 1201 performs some or all of the processing steps of the present disclosure in response to processor 1203, which executes one or more sequences of one or more instructions contained in memory, such as main memory 1204. Such instructions may be read from other computer-readable media, such as the hard disk 1207 or removable media drive 1208, into main memory 1204. One or more processors in a multi-processing arrangement may be used to execute the instruction sequence contained in main memory 1204. In alternative embodiments, wiring circuits may be used in place of or in combination with software instructions. Therefore, embodiments are not limited to any particular combination of hardware circuits and software.

As mentioned above, the computer system 1201 comprises at least one computer readable medium or memory containing instructions programmed according to the teachings of the present disclosure and containing data structures, tables, records or other data described herein. Examples of computer-readable media include compact disks, hard disks, floppy disks (registered trademarks), tapes, magneto-optical disks, PROMs (EPROM, EEPROM, flash EPROM), DRAM, SRAM, SRAM, and any other magnetic medium or compact disk. (Eg, CD-ROM), any other optical medium, punch card, paper tape, or other physical medium with a pattern of holes.

The present disclosure includes software that controls a computer system 1201, drives one or more devices to realize the invention, and allows the computer system 1201 to exchange information with a user. It is stored in any one or a combination of computer-readable media. Such software includes, but is not limited to, device drivers, operating systems and application software. Such a computer-readable medium further comprises the computer program product of the present disclosure for performing all or part (if the processing is distributed) of the processing performed in realizing the invention.

Computer code devices, any interpretable or executable code that includes, but is not limited to, scripts, interpretable programs, dynamic link libraries (DLLs), Java® classes and full executable programs. It may be a mechanism. In addition, some of the processing of this embodiment may be distributed for better execution, reliability and / or cost.

As used herein, the term "computer-readable medium" means any non-transitory medium involved in providing instructions to processor 1203 for execution. Computer-readable media may take many forms, including but not limited to non-volatile or volatile media. Non-volatile media include, for example, optical disks such as hard disks 1207 or removable media drives 1208, magnetic disks and magneto-optical disks. Volatile media include dynamic memory such as main memory 1204. In contrast, the transmission medium includes the wires that make up bus 1202, including coaxial cables, copper wires and optical fibers. The transmission medium may take the form of sound waves or light waves that occur during radio wave communication and infrared data communication.

Various forms of computer-readable media may also relate to the execution of one or more sequences of one or more instructions to processor 1203 for execution. For example, the instruction may be executed first on the magnetic disk of the remote computer. The remote computer can load instructions for remotely implementing all or part of the present disclosure in dynamic memory and transmit the instructions via a telephone line via a modem. A modem near computer system 1210 may receive telephone line data and place the data on bus 1202. Bus 1202 transmits data to main memory 1204, and processor 1203 reads and executes instructions from main memory 1204. The instruction received by the main memory 1204 may be arbitrarily stored in the storage device 1207 before or after the execution by the processor 1203.

Computer system 1201 also includes communication interface 1203 coupled to bus 1202. The communication interface 1213 provides bidirectional data communication coupled to a network link 1214 connected to, for example, a local area network (LAN) 1215 or another communication network 1216 such as the Internet. For example, the communication interface 1213 may be a network interface attached to an arbitrary packet switching LAN. As another example, the communication interface 1213 may be a service integrated services digital network (ISDN) card. A wireless link may be realized. In such an implementation, the communication interface 1213 transmits and receives electrical, electromagnetic or optical signals that transmit digital data streams representing various types of information.

Network link 1214 typically provides data communication to other data devices over one or more networks. For example, the network link 1214 may provide a connection to another computer through a device operated by a service provider that provides communication services through a local network 1215 (eg, LAN) or through a communication network 1216. The local network 1214 and the communication network 1216 use, for example, an electrical signal, an electromagnetic signal or an optical signal for transmitting a digital data stream, and a related physical layer (for example, CAT5 cable, coaxial cable, optical fiber, etc.). Signals through various networks that transmit digital data to and from computer system 1201 and signals on network link 1214 through communication interface 1213 are baseband signals or carrier wave based signals. ) May be realized. Baseband signals carry digital data, such as unmodulated electrical pulses that describe a stream of digital data bits, in which case the term "bit" should be broadly interpreted to mean a symbol. Each symbol carries at least one or more information bits. Digital data may be used, for example, to modulate a carrier wave with an amplitude, pulse and / or frequency shift keyed signal propagated through a conductive medium or transmitted through a propagation medium as an electromagnetic wave. Thus, digital data may be transmitted as unmodulated baseband data through a "wiring" communication channel and / or may be transmitted within a predetermined frequency band different from the baseband by modulating the carrier wave. Good. The computer system 1201 can send and receive data including program code through (one or more) networks 1215, 1216, network link 1214, and communication interface 1213. Further, the network link 1214 may provide a connection with an electronic device (eg, a mobile device) 1217 via the LAN 1215.

The various elements, forms and processes described herein may be used independently of each other or in combination in various ways. It is intended that all possible and partial combinations are within the scope of this disclosure. Further, in the present disclosure, it means that any particular form, element, component, characteristic, step, module, method, process, task or block is essential or essential (unless otherwise indicated). Not intended to be. Examples of the system and components described herein may be configured differently from those described. For example, the element or component may be added to the disclosed example, removed or rearranged from the disclosed example.

As used herein, reference to "one embodiment," "some embodiments," or "embodiments" is a particular element, form, structure, or characteristic described in connection with an embodiment. Means included in at least one embodiment. The appearance of the phrase "one embodiment" in various places in the specification does not necessarily refer to all the same embodiments. It is meant that all possible and partial combinations of embodiments are within the scope of the present disclosure. In particular, conditional terms used herein, such as "can," "may," "for example," etc., are generally defined unless otherwise specified or understood within the context in which they are used. It is intended to convey that one embodiment includes a predetermined form, element and / or step, whereas another embodiment does not include a predetermined form, element and / or step. Furthermore, "one" (articles "a" and "an") used in the present application and the appended claims means "one or more" or "at least one" unless otherwise specified. Should be interpreted as.

Claims (17)

A device that determines the angle of rotation of text in an image, with a circuit
The circuit
For each object of the plurality of objects contained in the text, the distance to the closest object is calculated.
Calculate the average distance to the closest object and calculate
The ratio between the average distance and the average font line width, which is the average of the font line widths of each of the plurality of objects, is determined.
A device configured to determine the angle of rotation of the text by comparing the ratio with a threshold. The circuit is further configured to determine the circumscribed rectangle of each object, which surrounds each object and touches the outermost part of each object.
The apparatus according to claim 1, wherein the circuit calculates the distance to the closest object by calculating the distance between each circumscribing rectangle and the closest circumscribing rectangle. The circuit
For each object, an object area corresponding to the entire print area of each object in the image is determined.
For each object, determine the object perimeter corresponding to the perimeter of each object, and
The apparatus according to claim 1, further configured to calculate the font line width of each object by dividing the object area by the outer circumference of the object and multiplying by 2. The apparatus according to claim 3, wherein the circuit is configured to calculate an average font line width by dividing the sum of the font line widths of the plurality of objects by the total number of the plurality of objects. The circuit is before calculating the distance to the closest object for each object.
Determine one or more text blocks of text in the image, each text block containing multiple objects.
For each text block, determine whether the text block is portrait or landscape, and
The device of claim 1, wherein the device of claim 1 is configured to calculate the distance to the closest object to each object of the text block in response to the determination that the text block is portrait orientation. For each object of multiple objects contained in the text in the image, calculating the distance to the closest object and
Calculating the average distance to the closest object and
Determining the ratio between the average distance and the average font line width, which is the average of the font line widths of each of the plurality of objects.
Determining the angle of rotation of the text by comparing the ratio with the threshold
How to prepare. Determining the circumscribed rectangle of each object, said circumscribed rectangle further comprising enclosing each object and contacting the outermost part of each object.
The method of claim 6, wherein calculating the distance to the closest object is performed by calculating the distance between each circumscribing rectangle and the closest object. For each object, determining an object area that corresponds to the entire print area of each object in the image.
For each object, determining the object perimeter corresponding to the perimeter of each object,
To calculate the font line width of each object by dividing the object area by the outer circumference of the object and multiplying by 2.
The method according to claim 6, further comprising. The method according to claim 8, further comprising calculating the average font line width by dividing the sum of the font line widths of the plurality of objects by the total number of the plurality of objects. Before calculating the distance to the closest object for each object,
Determining one or more text blocks of text in the image, each text block containing a plurality of objects.
For each text block, determining whether the text block is portrait or landscape
In response to the determination that the text block is portrait orientation, calculating the distance to the closest object to each object in the text block, and
The method according to claim 6, further comprising. The method of claim 6, further comprising performing optical character recognition (OCR) after determining the angle of rotation of the text. When running on a computer
For each object of multiple objects contained in the text in the image, calculating the distance to the closest object and
Calculating the average distance to the closest object and
Determining the ratio between the average distance and the average font line width, which is the average of the font line widths of each of the plurality of objects.
Determining the angle of rotation of the text by comparing the ratio with the threshold
A non-temporary computer-readable storage medium having computer-executable instructions that cause a computer to perform a method comprising. Determining the circumscribed rectangle of each object, said circumscribed rectangle further comprising enclosing each object and contacting the outermost part of each object.
The non-temporary computer-readable storage medium according to claim 12, wherein calculating the distance to the closest object is performed by calculating the distance between each circumscribing rectangle and the closest object. For each object, determining an object area that corresponds to the entire print area of each object in the image.
For each object, determining the object boundaries that correspond to the boundaries of each object,
To calculate the font line width of each object by dividing the object area by the object boundaries and multiplying by 2.
12. The non-temporary computer-readable storage medium according to claim 12. The non-temporary computer-readable storage medium according to claim 14, further comprising calculating the average font line width by dividing the sum of the font line widths of the plurality of objects by the total number of the plurality of objects. Before calculating the distance to the closest object for each object,
Determining one or more text blocks of text in the image, each text block containing a plurality of objects.
For each text block, determining whether the text block is portrait or landscape
In response to the determination that the text block is portrait orientation, calculating the distance to the closest object to each object in the text block, and
12. The non-temporary computer-readable storage medium according to claim 12. The non-temporary computer-readable storage medium according to claim 12, further comprising performing optical character recognition (OCR) after determining the angle of rotation of the text.

Images