 |
Description  |
|
|
BACKGROUND OF THE INVENTION
This invention relates in general to character recognition apparatus. More
specifically, the present invention provides a character recognition
apparatus that operates substantially in real time to identify characters
drawn using a stylus applied to a tablet. The characters can include
alpha-numeric characters, marks, and the like.
Many attempts have been made to automatically recognize handwritten
characters drawn using a tablet/stylus pen combination for indicating the
relative position of the stylus pen with respect to the tablet. Many of
such attempts have included the use of complex software for computer-aided
analyzing of the characters. However, there are significant disadvantages
attendant this approach. the computer software necessary for character
recognition programs is extremely complex. It takes a long time to
recognize any given character. This means that character recognition
cannot be carried out in real time which significantly limits the
application of such a program. It cannot be used "on line" without causing
a user to be frustrated.
An alternative to this type of computer-aided recognition is to recognize a
character by classifying various typical elements of the character and
comparing them with elements predefined in a "look-up" table of elements
of known characters. This approach for character recognition can be used
when the character can be easily represented by such typical elements
which are composed of combinations of straight lines. For such characters,
relatively high accuracy of recognition can be achieved.
SUMMARY OF THE INVENTION
The present invention provides an apparatus for recognizing characters. The
character can be an alpha-numeric character or mark. Recognition occurs
substantially in real time as the character is written using a stylus pen
on a surface of a tablet which is provided with a pen-position detector.
During movement of the pen over the tablet surface, initiation and
completion of a stroke of a character is detected when the vertical
position of the pen which is expressed as the Z-coordinate shifts. The end
of writing a character is also detected either when the pen has moved
across the tablet far beyond the distance of a normalizing size of the
character or when the passage of time between the occurrence of two
adjacent discontinuities of the Z-coordinate exceeds a preset value, e.g.,
when the stylus is lifted off the tablet for more than a predetermined
time.
The length (L) and the gradient (.theta.) of a line segment of a stroke is
determined by examining the X and Y coordinates of the start and end
points of the line segment. The line segments are classified into plural
categories and designation codes are assigned to each line length and
gradient. The particular order of occurrence of the codes identifies
stroke order and hence a character. A specific character is determined by
comparing the code order with those preset in a table of codes.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a block diagram of the character recognition apparatus according
to the present invention;
FIG. 2 graphically illustrates the assignment of a code for each of various
line segment directions of a stroke of a character;
FIG. 3 illustrates how a character such as the numeral character "4" is
composed of distinct strokes;
FIGS. 4(a) and 4(b) illustrate the step for characterizing line segments
and strokes of a character according to the present invention; and
FIG. 5 illustrates an analysis of the transition of characteristic codes of
strokes of characters to be recognized.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT
Referring to FIG. 1, there is shown a block diagram of the character
recognition apparatus according to the present invention. A character,
which can be an alpha-numeric character or some other mark, is described
by moving a stylus pen 3 over a surface of a tablet 1. The stylus pen 3
and tablet are selected such that either or both provide information as to
the relative position of the stylus and tablet such as, for example, in an
XYZ coordinate system of the tablet (the Z-coordinate meaning whether the
stylus pen 3 is in contact with tablet 1 or is lifted from it). Onc
suitable tablet for this purpose is one supplied by Daini-Seikosha type
DT-4000. Of course, the present invention is not limited to this
particular type of tablet. Rather, the invention depends only upon the
tablet 1/stylus pen 3 combination providing X, Y and Z information. In
order to assist the user, a piece of paper 2 is attached to the writing
surface of tablet 1 so that the user can see what he is drawing. As shown
in the Figure, the user is drawing the numeral "4".
Outputs from the stylus pen 3/tablet 1 are coupled to an X-Y reading
circuit 4. X-Y reading circuit 4 establishes the position of stylus pen 3
with respect to the surface of tablet 1. An output of X-Y reading circuit
4 is coupled to a character completion detector 5 for determining when the
drawing of once character has been completed and the drawing of the next
character is begun. This is done in two ways. First, a character is
assumed to have a normalizing size. If stylus pen 3 is moved from the end
point of one line segment further than a normal distance consistent with
the character size, it is assumed that the user is moving onto a new
character. Secondly, time is measured from the time the stylus pen 3 is
lifted from the surface of tablet 1 until the stylus pen is replaced on
the surface of the tablet. For strokes in a single character, this time
will not be excessively long. However, this time is longer between the end
stroke of a first character and the beginning stroke of a second.
Therefore, a measurement of the time period also provides an indication of
the beginning of a new character.
An output of character completion detector 5 is coupled to an X-Y
coordinate memory 6. X-Y coordinate memory 6 memorizes positions of stylus
pen 3 by storing its three coordinates XYZ from a time when stylus pen 3
touches paper 2 (Z=1) to a time when the stylus pen is lifted from the
surface of the paper (Z=0). It is determined that a character has been
written when the start of a line segment is located more than a
predetermined distance on either coordinate X or coordinate Y when stylus
pen 3 is on the paper (Z=1) from the end of the previous line segment. For
example, it is determined that a character has been written when the pen
has been shifted to a position 20 mm on either coordinate X or coordinate
Y from its former position for the case wherein the character sizes 10
mm.times.10 mm.
An output of X-Y coordinate memory 6 is coupled to an Xmax, Ymax detection
circuit 7. This circuit detects minimum and maximum values of X and Y as
accumulated in X-Y memory 6. An output of Xmax, Ymax detection circuit 7
is coupled to a character size calculation circuit 8. Character size
calculation circuit 8 calculates the difference between the maximum
X-value (Xmax) and the minimum X-value (Xmin) as a value A. It also
calculates the difference between a maximum Y-value (Ymax) and a minimum
Y-value (Ymin) as a parameter B. Thus, A and B are expressed by equations
1 and 2.
A=Xmax-Xmin (1)
B=Ymax-Ymin (2)
The size of the character is represented as A.times.B, each corresponding
to the X and Y coordinates.
An output of character size calculation circuit 8 is coupled to a
normalizing circuit 9. Normalizing circuit 9 normalizes the size of a
character by transformation of the X and Y coordinates. For example, a
written character can be normalized into the size of C x C corresponding
to each of the X and Y coordinate axes by the following equations:
X'=C (X.sub.t -Xmin).div.A (3)
Y'=C (Y.sub.t -Ymin).div.B (4)
where X'.sub.t and Y'.sub.t are transformation coordinates and the written
original position is expressed by X.sub.t and Y.sub.t on the coordinate
axes X and Y, respectively. Thus, a written character is normalized in
size even when the character size varies from time to time. In essence,
the absolute coordinates for any character in memory 6 are transformed to
relative values. Normalizing circuit 9 also receives information directly
from X-Y memory 6 in addition to the information processed by Xmax, Ymax
detection circuit 7 and character size calculation circuit 8.
Normalizing circuit 9 provides information to a register Z10, a register
X13 and a register Y16. The (t) shown in each of these three blocks in
FIG. 1 represents a time t. Registers 10, 13 and 16 store the normalized
or transformed values for a character as determined by normalizing circuit
9. These three register transmit their respective contents at a time t-1
to a register Z11, a register X14 and a register Y17, respectively. The
FIG. 1 blocks for registers 11, 14 and 17 include a t-(1) to indicate the
transfer information from registers 13 and 16 at time t-1. This transfer
of information from registers 10, 13 and 16 to registers 11, 14 and 17,
respectively, occurs just before new information is received from
normalizing circuit 9 and stored in registers 10, 13 and 16 for the next
character.
A Z differential circuit 12 is coupled to the outputs of registers 10 and
11 and determines whether stylus pen 3 position has been changed in the
Z-coordinate by comparing the values from registers 10 and 11. It
determines, for example, whether there has occurred a shift of the
Z-coordinate from Z=1 to Z=0 or from Z=0 to Z=1 or whether the coordinate
remains unchanged.
An X differential circuit 15 is coupled to outputs of X registers 13 and
14. X differential circuit 15 determines the change of X-coordinate
between two points in a character in accordance with equation 5:
a=X'.sub.t -X'.sub.t-1 (5)
where X'.sub.t represents the content of register 13 and X'.sub.t- 1
represents the content of register 14.
A Y differential circuit 18 is coupled to outputs of Y registers 16 and 17.
It determines any change b of the Y-coordinate value as expressed by
equation 6:
b=Y'.sub.t -Y'.sub.t-1 (6)
where Y'.sub.t represents the content of register 16 and Y'.sub.t-1
represents the content of register 17.
A line direction detecting circuit 19 is coupled to the output of Y
differential circuit 18 and X differential circuit 15. Line direction
detecting circuit 19 determines the line direction between two positions
in one stroke of a character. In other words it decides the direction of
movement of stylus pen 3 along the surface of tablet 1. The gradient
.theta. or slope of the line between these two adjacent positions is
determined by equation 7:
tan.theta.=b.div.a (7)
For the purpose of simplifying recognition steps, .theta. is classified,
for example, into eight consecutive codes as shown in FIG. 2. The
invention is not limited to the use of eight codes corresponding to eight
directions but is only shown in the manner for simplicity. For greater
resolution, the directions can be further divided into, for example, 16 or
36 different directions with corresponding 16 or 36 different codes.
Once line direction has been determined by direction detector 19, a signal
carrying this information is coupled to a register 20. Register 20
provides this information in turn to a line direction differentiator 24 at
a time just before it receives new .theta. information from direction
detector 19. Direction differential circuit 24 is also coupled directly to
the output of direction detector 19. Thus, direction differential circuit
24 can determine changes in gradient .theta. occurred during the "delay
time" of register 20.
A line length detector 21 is provided with inputs from X differential
circuit 15, Y differential circuit 18 and Z differential circuit 12.
Detector 21 calculates line length L in accordance with equation 8:
##EQU1##
Line length detector 21 provides a signal indicative of length L to a
comparator 22 which also receives a signal from character size calculation
circuit 8 indicative of character size. Comparator 22 in effect neglects
any movement of stylus pen 3 when L is less than a predetermined value E
which predetermined value is related to the normalized size of the
character C under consideration. In such a case when the movement of the
pen is neglected, Z-coordinate register 11, X-coordinate register 14 and
Y-coordinate register 17 are kept unchanged. The normalized X, Y and Z
coordinates observed at the next time of line position sampling are sent
to registers 10 (Z), 13 (X) and 16 (Y). However, when line length L is
larger than the predetermined value of E, it is assumed that a line
segment is being written. The line length information L is coupled to a
length adder 23 in the form of a signal from comparator 22 to the length
adder 23. As stated above, direction differential circuit 24 determines in
change in .theta. over a period of time defined by a delay through
register 20. This information is coupled via a signal from direction
differential circuit 24 to length adder 23. When .theta. is found to be
unchanged, the signal to length adder 23 in effect commands length adder
to add the length L of the line segment.
However, when a change of .theta. is observed, the .theta. data is provided
to adder 23. The added length data is coupled from length adder 23 to a
length classification circuit 26 while line length data L is coupled from
comparator 22 to length adder 23. Therefore, as stylus pen 3 moves across
the surface of tablet 1, length L accumulates until there is an indication
of a change in gradient .theta.. At that point, the length of the line
segment is fixed and length L begins to accumulate for the next line
segment at a new gradient. Line length classification circuit 26
classifies line length data L from adder 23 into one of plural categories.
For example, a quarter of the normalized character size C may be used as a
measure for such categories. For higher resolution, one could divide the
normalized character size into more parts.
The process of extracting information as to the distinction of a stroke of
a character will now be described. A stroke of the character begins when
stylus pen 3 touches the surface (Z=1) of tablet 1. The stroke continues
until stylus pen 3 leaves the surface of the tablet (Z=0). The circuitry
for distinguishing a character includes a memory 27, preferably a read
only memory (ROM), and a register 28. ROM 27 receives information from
register 20 and from length classification circuit 26 and provides an
output to register 28. Register 28 also receives an input from direction
differential circuit 14 through an AND gate 29 which is clocked by a clock
signal CK1. Register 28 provides a feedback to an input of ROM 27. Signals
are coupled to register 28 in accordance with an output timing of gate 29
in accordance with the clock signal CK1 applied to it. Thus, register 28
receives information from ROM 27 when there is a coincidence of clock
signal CK1 and an output from line direction differential circuit 24.
A stroke distinction signal is outputted when line direction changes. As
information flows from ROM 27 to register 28, changes of line direction
are stored in register 28 and the register is cleared after its contents
have been transferred to a further read only memory 30. Read only memory
30 receives a signal from a relative position detector 25. Relative
position detector 25 processes information from Z differential circuit 12,
X differential circuit 15 and Y differential circuit 18. An output of ROM
30 is coupled to a further register 31 which provides the ultimate output
signal of the character recognition device and feeds back a signal to ROM
30. Register 31 is operated by a signal from Z differential circuit 12
coupled through a gate 32 which is clocked by a clock signal CK2.
To further explain the operation of the character recognition apparatus
according to the present invention, an example will be discussed. In this
example, the numeral "4" is written by stylus pen 3 as illustrated by the
marks written on paper 2 as shown in FIG. 1.
Referring now to FIG. 3, there is shown a numeral "4" which will be written
by stylus pen on the surface of tablet 1. This numeral has component parts
as will be illustrated in FIG. 4. For this example, the resolution of line
direction .theta. is classified into eight units or categories as shown in
FIG. 2. Line length L is classified into four categories including 0, 1, 2
and 3 for a normalized character size C. The relation between length code
and absolute line length is set forth in the following table:
______________________________________
length code absolute line length
______________________________________
0 L .ltoreq. 0.25 C
1 0.25 C < L .ltoreq. 0.5 C
2 0.5 C < L .ltoreq. 0.75 C
3 0.75 C < L
______________________________________
The initial stroke L shown in FIG. 4(a) constitutes the first straight line
segment drawn and forming the numeral character "4". It has a gradient
code .theta.=5 by reference to FIG. 2 and a length code L=3. Continuing to
refer to FIG. 4(a), the second line segment which is substantially
horizontal, has a gradient code .theta.=0 and a length code L=3. These two
line segments constitute the first stroke in forming a numeral character
"4". Referring now to FIG. 4(b)there is shown the second stroke in forming
the numeral character "4". This stroke consists of only a single straight
line segment having a gradient .theta.=6 and a length L=2.
Referring now to FIG. 5, there is illustrated the process by which a
particular character is distinguished from other characters based upon the
strokes established such as was done in the FIG. 4 example.
For instance, assume a starting position C00. A first line segment having
.theta.=5 and L=3 would produce distinction C02. Once having made
distinction C02, a further line segment having .theta.=7 and L=3 would
produce distinction C09. However, from distinction C02, a further line
segment having .theta.=0 and L=3 would produce distinction C04. As you can
see from FIG. 5, distinctions C04 and C09 are quite different thereby
identifying two different characters. Returning to starting point C00, a
first line segment having .theta.=6 and L=3 would produce distinction C01.
A further line segment having .theta.=0 and L=3 would produce distinctions
C03. Returning to starting point C00, a first line segment having
.theta.=0 and L=3 would produce distinction C05. Once having made
distinction C05, a further line segment of .theta.=6 and L=3 would produce
distinction C06. However, from distinction C05, a further line segment of
.theta.=5 and L=3 would produce distinction C07. After having made
distinction C07, a further line segment of .theta.=0 and L=3 would produce
distinction C08. Thus, a first line segment in accordance with distinction
C05 could lead to any of the characters represented by distinctions C06,
C07 and C08.
To further explain the process, refer back to distinction C02 which occurs
after a line segment .theta.=5, L=3. Then, see C02 changes to C04 when a
line having code .theta.=0 and L=3 is encountered. The second distinction
C04 is registered and the final distinction of that stroke is decided as
C04 if Z becomes 0 before the third distinction appears. The second stroke
of the numeral "4" (as shown in FIG. 4(b)) is one straight line having a
designation code .theta.=6 and L=2. The distinction for the second stroke
of the numeral "4" is C01.
Referring back to FIG. 1, the character recognition is processed by read
only memory 30 and register 31. The output signal of the Z-coordinate,
generated upon a change from 1 to 0, and a clock signal CK2 are applied to
AND gate 32.
Considering again the example of the numeral "4" as shown in FIG. 3, the
first stroke corresponds to distinction C04 shown in FIG. 5 and the code
corresponding to C04 would be inputted to ROM 30 and ultimately set in
register 31.
The second stroke distinction C01 and the first readout data, now in
register 31 would be combined. The results of that combination of two
distinctions of strokes of the numeral "4" would be the combination of
distinctions C01 shown in FIG. 5 and C04 shown in FIG. 5. This would
result in the final recognition of the character as numeral "4".
In the character recognition apparatus, character completion detector 5
(see FIG. 1) detects completion of one character when X and Y coordinates
deviate beyond predetermined value E when Z=1. The completion of a
character can also be detected by measuring the passage of time during the
state of Z=0 (stylus off of the tablet). Using this alternative, the
completion of a character is detected when the duration for Z=0 exceeds a
predetermined time.
The normalization of character size can also be achieved by normalizing the
length of segment lines of strokes of a character after the length and
directions of the line segments are calculated. This is in effect an
alternative to calculating length and direction after transforming
coordinates of lines. The process of extraction of the distinction of
stroke and the later recognition of the ultimate character being drawn
based on the extracted distinction can be modified to accomplish the same
objective by processing all data related to direction and length of
segment lines of the strokes within one complete character after having
been stored in memory. The procedure of extracting character distinction
and/or character recognition can be accomplished by a microcomputer or
some special purpose electronic device other than the combination of ROM
30 and register 31 shown in FIG. 1. Using the character recognition
apparatus of the present invention, the character is recognized
substantially in real time without utilizing complicated software
programs.
While the invention has been described in connection with what is presently
considered to be the most practical and preferred embodiments, it is to be
understood that the invention is not to be limited to the disclosed
embodiments but on the contrary is intended to cover various modifications
and equivalent arrangement included within the spirit and scope of the
appended claims which scope is to be accorded the broadest interpretation
so as to encompass all such modifications and equivalent structures.
* * * * *
|
|
 |
|
 |
Description |
|
