JPS6054085A - Aloud reading type character reader for visually handicapped person - Google Patents

Abstract

PURPOSE:To enable to read any part by cheap constitution by reading characters to be read optically and successively, storing it once and making recognition, and outputting a sentence as a series of voices basing on the result of recognition. CONSTITUTION:An area sensor is structured to be moved to any position on the character face by manual operation and to detect a picture of small area optically. Output of the area sensor is taken in a present sample picture memory. Then, degree of superposition of two pictures is calculated to stick with previous sample picture stored in a previous sample picture memory, and the two pictures are stuck together and transferred to a buffer. Then, a mode controller selects operation mode by manual switching in a recognizing section. A recognizing device in the recognizing section cuts out a picture signal for one letter from continuous group of pictures and recognizes the character. An outputting section outputs a sentence in which distinction of phonetical reading and Japanese reading is given to each character and reading of characters appointed by the area sensor as voices.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an aloud character reader for visually impaired people that transmits text information by voice to severely visually impaired people.

As a means of transmitting text information to severely visually impaired people,
Braille is generally used, and the amount of information expressed in Braille is almost nothing compared to printed matter using ordinary characters. In addition, existing optical character readers that can process general Japanese information are minicomputer-based automatic scanning page types, which not only lack the freedom to read the parts you want to read, but are also expensive and difficult for individuals to read. It's impossible to own it.

In view of the foregoing, it is an object of the present invention to construct a device that reads characters at an arbitrary location by manual operation and outputs a voice as simple and inexpensive as a microcomputer.

In order to achieve the above object, the audible character reader of the present invention includes an input section, a recognition section, and an output section, and the input section can be manually operated to move to any position on the character surface to read small areas. An area sensor that optically detects the image of
A y counter, a current sample image memory that can be connected to the area sensor via a binarization circuit, a previous sample image memory that stores the current sample image, and the degree of overlap between the two images in order to paste them together. The residual calculation circuit calculates the number of black pixels of the current sample image as η−
The recognition unit is equipped with a mode control device that switches and selects the operation mode, and extracts image signals for each character from a continuous image group (2). and a recognition device that recognizes the characters by comparing the result with the characteristics of various characters stored in advance as a dictionary in the memory, and outputs sounds 1i111 i1', fl! for each character in the output section. The four characteristics include an output device that outputs audio signals of all the readings of the text specified by the area sensor, and a speaker that outputs audio based on the audio signals. It is something.

According to the character reader of the present invention, the area sensor for reading characters is configured to be operated manually, so that a user can read any part repeatedly at a speed that suits him/her as many times as desired. Moreover, since the structure is simple and based on a microcombinator, the product can be obtained at low cost.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

As shown in FIG. 1, the aloud character reader according to the present invention includes a human power section, an 88 section, and an output section. The characters are sequentially optically read by the part and stored once, the recognition part recognizes those characters, and based on the recognition result, the output part outputs a sentence made up of those characters as a series of sounds. It consists of

To explain more specifically, the area sensor in the input section converts the image of the character on the target surface into two using a lens system.
It focuses an image on the light-receiving surface of a dimensional sensor array, etc., and converts the resulting two-dimensional image into a matrix array of electrical signals. The area sensor optically reads characters in a small area covered by the area sensor (operated twice).

As shown in FIG. 2, this area sensor 1 is movable in two directions, x and y, which are directly connected to each other on a two-dimensional plane in order to scan the character surface; The first frame 2 is restrained by the second frame 3 and moves one-dimensionally in the X direction on its 7 frames 2, and the first frame 2 is restrained by the second frame 3 and moves one-dimensionally in the y direction. It is configured as something that does The dimensional position (x, y) of the area sensor 102 that moves in this way is determined by the position X of the area sensor 1 with respect to the first frame 2 and the position Y of the first frame 2 with respect to the second frame 3. As a means of determining the above position (x, y),
A method of detecting by attaching a scale to the first and second frames 2.3 and attaching a sensor composed of an infrared light emitting source and a photodiode to the area sensor 1 and the first frame 2, or a general-purpose digitizer. It is possible to adopt a method of detecting coordinates using . The position of this area sensor l is displayed to the recognition unit by the xy counter shown in FIG.

The input section includes a current sample image memory, a previous sample image memory, a buffer, a residual R1 calculation circuit, an addition projection register, etc., which are configured with a binarization circuit and a frame memory connected to the area sensor. ing.

The current sample image memory is connected to the above area sensor via a binarization circuit, and the above electric signal group output from the area sensor is converted into a sample image signal by the binarization process in the binarization circuit. The current sample is loaded into image memory.

Next, the position of the sample image imported into the current sample image memory in this manner relative to the previous sample image stored in the previous sample image memory is determined. This paste is used to determine the position of the current sample image and the previous sample image so that they continue in a proper state. A calculation circuit calculates the degree of local overlap (residual difference) between both images, determines the position where it is minimum, and pastes both images together at that position and transfers them to a buffer.

To explain more specifically about pasting both images,
First, when pasting both images together in the X direction, the area sensor is configured as an MxN mesh as shown in Figure 6, and sampling is performed at a position of '1', 11', and 1 mesh in the X direction. For example, p 1q +T+ = N where: 9: The error range of the position to be considered is P2N, which is the evaluation range. Therefore, the N-7th image in the previous frame image B)
Find the minimum α by taking the sum of the exclusive ORs from the 1,001st column to the Nth column and from the αth column to the p10th α column (α moves from 0 to 29) in the current frame image A, α=q
If it becomes the minimum at o, both images A, 13Fip+9
They are connected with an overlap of ° and transferred to the buffer.

On the other hand, in order to perform positioning in the y direction, position sampling is performed in the y direction at intervals of 1 and 2 meshes, and in the line detection mode described later, the first frame 2 in Fig. 2 is moved by the necessary amount. At this time, a stop signal in the form of an audio display is fed back using information on segmentation (described later) in the recognition unit. ′ The addition projection register connected to the current sample image memory calculates the number of black pixels of the current sample image in the area covered by the area sensor on the character surface, determines the grade, transfers it to the output unit, and outputs the audio according to the grade. By displaying the information, the character surface situation at any position is fed back.

That is, based on the image signal of an arbitrary position covered by the area sensor, the recognition unit determines whether a character part is detected or not, and the result is outputted as a sound from the output unit. The reference value for line detection is calculated in the recognition unit from the amount of movement of the area sensor required until the next character part is detected again after a character part is detected as described above.

The mode control device in the recognition section disposed next to the input section allows the following four operating modes to be selected by manual switching.

(Re) A text search mode for finding where the printed text is located on the text surface of the book you are trying to read.

(2) Line determination mode in which you search for the next line after reading one line of characters.

(3) A skim reading mode that allows you to read one line of text continuously, separating them with punctuation marks.

(4) An explanation mode that outputs all readings of a character in order to deepen understanding of the character when a character in a text cannot be understood well, such as when skimming mode is adopted.

7) By switching this mode, each key point in Figure 1 can be adjusted.
Their driving is controlled to selectively operate the elements.

The recognition device in the recognition unit cuts out an image signal for each character from a continuous image group detected by an area sensor moving on a character string, and also identifies what kind of character the cut out image signal is. It is equipped with a simple algorithm that recognizes and outputs a character from a dictionary using a dictionary, and performs segmentation that cuts out an image signal for one character from the above series of image signals using the projective addition method, and a character that is composed of the extracted image signals. It has a feature calculation function that first calculates the external shape feature and then a linear density feature for the character, and a matching function that uses the calculation results to determine which character the character corresponds to previously stored as a dictionary in memory. It is constructed as a thing. Calculation of the above external shape features is performed using shift registers, logical registers, projection addition registers, etc.
As shown in Figure 4, the shadow part of the character line is replaced with black pixels from any of the four directions (top, bottom, left, right), and the area of the character is equally divided into 16 small areas of 4 x 4. Black pixels (“10”) or white pixels (
A 2-bit pattern of "01") is assigned to represent the - character with a 32-bit pattern.

In order to select and extract candidate characters based on such external shape percentage characteristics, various characters and corresponding bit patterns similar to the above are stored as a dictionary in the memo ↓ above, that is, in 16 small areas. Black pixel (“10”), white pixel (“01”)
), and for small areas that are unstable due to the font (“11”
) is stored in memory. Then, from the characters in the dictionary, detect a character group represented by a pattern AI that satisfies the following: At A2 = A2 (: logical product) Al: pattern of a certain character in the dictionary A2: unknown character pattern,
These characters are set as candidate characters based on the external shape percentage characteristics.

Next, in order to select one character from among the candidate characters as a proper character, a comparison is performed using the line density feature. This linear density percentage is calculated using a shift register, a projection addition register, etc. In the same way as above, first, the image of an unknown character is divided into 16 small areas, and the horizontal and vertical The total number of pixel state changes in each direction is calculated, and the resulting 32-dimensional vector is used to calculate the similarity with the average 32-dimensional vector pattern across various fonts in the dictionary using an array processor. It outputs similar characters as search results.

The above output section is for outputting in audio whether or not the area sensor has detected a character part or the next line, or the sentence traced by the above area sensor or the pronunciation of a single character in the sentence. It is constructed by connecting an output device for performing exception processing, rule processing, and voice synthesis, and a speaker to an output buffer connected to a control device.

The above exception processing outputs the corresponding pronunciation for a character string that matches a certain compound word that has been registered in the compound word dictionary in advance,
In addition, according to the registration results of On-yomi and Kun-yomi, the rule processing outputs the On-yomi when the kanji are consecutive, and the W11 reading when they are alone.Furthermore, when the explanation mode is adopted, the kanji registered in the dictionary of the rule-processing All readings are output. Furthermore, when the digit determination mode or the literature search mode is adopted, a sound is output to notify whether the next line or character section has been detected. Speech synthesis is based on the signal output by the above-mentioned process, and converts the human monosyllabic speech waveform into A7. The audio signal is artificially synthesized using a monosyllabic editing synthesis method that converts and samples the audio signal.The audio signal is converted into audio by the next stage speaker, and whether or not a character part or the next line is detected is detected. You can also get the reading of a sentence or a single character audibly.

In the device configured as described above, if the mode control device selects the υ literature section search mode and moves the area sensor vertically and horizontally on the text surface of a book, etc., the area sensor will search for the text section. If you select skim reading mode and move the area sensor along the string of one digit, the moment you search,
The letters are output as a series of sounds, separated by punctuation marks. - After reading the characters of the gender, switch to line detection mode by the mode control device, move the area sensor within the next line, and when the area sensor reaches the next launch, the detection of that line will start. The text is output as audio from the speaker, and then each line can be played and read in sequence in the same way as above. For example, in the process, if the mode is switched to explanation mode, all the pronunciations of a specific character in the text can be obtained as audio from the speaker, thereby making it possible to reliably understand that character.

[Brief explanation of the drawings] Fig. 1 is an overall configuration diagram of an embodiment of the present invention, Fig. 2 is an explanatory diagram regarding the movement of the area sensor, and Fig. 6 is a combination of images continuously detected by the area sensor. FIG. 4 is an explanatory diagram of the state and the external shape characteristics of an unknown character detected by the area sensor. 1-...Area sensor.゛Designated Agent: Institute of Industrial Science and Technology, Director of Product Wood Science Research Institute, Keiji Takahashi Figure 2 Figure 3

Claims (1)

[Claims] 1. Consists of an input section, an output section, and an output section, wherein the input section is moved to an arbitrary position on the character surface by manual operation to optically detect an image of a small area. an area sensor, an xy counter that detects and outputs the position of the area sensor on the character surface, a current sample image memory connected to the area sensor via a binarization circuit, and an image captured in the current sample image memory. A previous sample image memory that stores the sample image immediately before the image, a residual calculation circuit that calculates the degree of overlap between the two images for pasting them together, and an addition projection register that calculates the number of black pixels in the current sample image. The recognition section is equipped with a mode control device for switching and selecting an operation mode, and a mode control device for cutting out an image signal for each character from a continuous image group, and storing the character feature calculation results in advance as a dictionary in the memory. It is equipped with a recognition device that recognizes the above-mentioned characters by comparing them with the memorized characteristics of various characters, and the above-mentioned output section is equipped with a recognition device that recognizes the above-mentioned characters by comparing them with the memorized characteristics of various characters, and the above-mentioned output section is provided with a sentence in which each character has been subjected to phonetic identification, and all of the characters specified by the above-mentioned area sensor. 1. An aloud character reader for visually impaired persons, comprising: an output device that outputs each reading as an audio signal; and a speaker that outputs audio based on the audio signal.