 |
Description  |
|
|
CROSS-REFERENCES OF THE RELATED APPLICATIONS
This application relates to an application U.S. Ser. No. 716,944 filed Mar.
28, 1985, entitled "Apparatus for Recognizing and Displaying Handwritten
Characters and Figures" by Fukunaga et al and assigned to the present
assignee, based on Japanese patent application No. 59-60717 filed Mar. 30,
1984, and whose disclosure is incorporated herein by reference.
BACKGROUND OF THE INVENTION
The present invention relates to a hand-written character/graphic
recognizer, and more particularly to a method for designating a
recognition mode necessary to separate hand-written characters/graphics.
In Japanese Patent Application Laid-Open No. 55-143681 entitled
"Hand-writing Word Processor Using a Tablet" (laid open Nov. 10, 1980),
function keys are provided to designate a recognition mode of
characters/graphics graphics so that an operator depresses them by a
stylus. In this designation method, the function key must be depressed
whenever a hand-written character/graphic is to be entered and hence it
has a low non-machine interface.
Japanese patent publication No. 58-42513 entitled "Real-time Hand-written
Character Recognizer" (published Sept. 20, 1983) teaches the designation
of recognition modes in the character recognition, such as alphanumeric
mode, Katakana mode, Kanji mode and Hirakana mode. In this recognizer, the
number of input strokes is examined and if it is larger than a
predetermined number, the recognition mode is automatically switched to
the Kanji/Hirakana mode.
Japanese Patent Application Laid-Open No. 57-105084 entitled "Method for
Separating Characters from Graphics" (laid open June 30, 1982) discloses a
method for recognizing characters separately from graphics. A drawing
containing both characters and graphics is read by a scanner and the read
image is repeatedly enlarged or reduced to separate the characters from
the graphics in an off-line method. It relates to an off-line recognition
in which characters and graphics are written or drawn on a sheet which is
then scanned for recognition. It cannot be applied to an on-line
hand-written character/graphic recognition.
Thus, in the past, there has been no effective method for automatically
designating the recognition mode for the characters/graphics in the
hand-written character/graphic recognizer. When a drawing containing
characters is to be prepared, the operator must designate the recognition
mode and the man-machine interface is not excellent.
SUMMARY OF THE INVENTION
It is an object of the present invention to provide a method for
automatically designating a recognition mode of a hand-written
character/pattern recognizer.
In accordance with the present invention, based on the finding that strokes
of a hand-written character are written in small rectangular areas,
characteristic values for recognizing the character/graphic are extracted
from stroke coordinates and the character recognition mode or the graphic
recognition mode is automatically designated in accordance with the
characteristic values. Major characteristic values are a maximum one of
projection length of the stroke coordinate on x-axis and y-axis, and a
stroke length.
In the present invention, the characteristic values are extracted each time
the stroke information is inputted.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is an overall block diagram of one embodiment of an on-line
hand-written character/graphic recognizer of the present invention.
FIG. 2 shows an example of hand-written strokes.
FIG. 3 shows an input order of the hand-written strokes shown in FIG. 2.
FIG. 4 illustrates extraction of a characteristic value from stroke
coordinates.
FIG. 5 shows projection lengths of all of the hand-written strokes shown in
FIG. 2.
FIG. 6 shows a status transition of recognition modes.
FIG. 7 shows a processing flow chart for mode recognition unit 3 shown in
FIG. 1 which is a characteristic of the present invention.
FIG. 8 shows a flow chart for minimum/maximum calculation in a step 3C of
FIG. 7.
FIG. 9 shows a sub-flow chart of handwritten character/graphic recognition
4 in FIG. 1.
FIG. 10 is a stroke table for a buffer used in the present invention.
FIG. 11 shows an example of hand-writte graphic strokes.
FIG. 12 illustrates a method for recognizing the hand-written graphic
strokes shown in FIG. 11 as graphic strokes.
FIG. 13 shows a processing flow chart when the method of FIG. 12 is applied
to FIG. 7.
FIG. 14 is a sub-flow chart when the method of FIG. 12 is applied to FIG.
9.
FIG. 15 is a processing flow chart in an embodiment in which a stroke
length is used as a characteristic value.
FIG. 16 illustrates how to determine the stroke length.
FIG. 17 shows an embodiment of an input/display device.
FIG. 18 is a processing flow chart corresponding to that shown in FIG. 7,
when the input/display device of FIG. 17 is used.
DESCRIPTION OF THE PREFERRED EMBODIMENTS
Referring to FIG. 1, numeral 1 denotes an input/display device which
includes input means 1a for detecting and inputting hand-written
character/graphic coordinates and display means 1b for displaying
characters/graphics. For example, the input means may be an
electromagnetic induction tablet and the display means may be a liquid
crystal display device. The input/display device may be constructed by
applying a transparent tablet onto the liquid crystal display device.
Numeral 2 denotes a stylus for hand-writing input. It also provides
touch-up and touch-down information of the stylus.
Numeral 6 denotes a character/graphic recognizer which includes a mode
recognition unit 3 for selecting a character or graphic mode, a
hand-written character/graphic recognition unit 4 and a character/graphic
display unit 5 for displaying characters/graphics on the display means 1b.
It may be a microprocessor.
The operation is now briefly explained. The stroke information written on
the input means 1a by the stylus 2 is inputted to the mode recognition
unit 3 in which a characteristic value is extracted from the stroke
coordinates included in the stroke information, the character or graphic
recognition mode is selected based on the extracted characteristic value,
and the hand-written character or graphic is recognized in the selected
recognition mode and the recognition result is displayed on the display 1b
through the character/graphic display unit 5.
The characteristic feature of the present invention resides in the mode
recognition unit 3 which is now explained in detail.
FIG. 2 shows an example of hand-written strokes. Characters " " are
written in a rectangle. In order to facilitate entry of the hand-written
characters/graphics, the input means are sectioned by a predetermined size
section (.DELTA.x x .DELTA.y). The small rectangular regions are
designated by numbers Nos. 1-15.
When such hand-written characters and graphics are entered by the strokes
S.sub.1 -S.sub.13 shown in FIG. 3, it is not clear which strokes are
character strokes and which strokes are graphic strokes.
In the present invention, a characteristics value as shown in FIG. 4 is
determined based on the stroke coordinates. For example, a projection
length .DELTA.u.sub.1 on the x-axis and a projection length .DELTA.v.sub.1
on the y-axis are determined based on coordinates P.sub.0 -P.sub.4 of the
stroke S.sub.1 and a longer one of .DELTA.U.sub.1 and .DELTA.V.sub.1 is
selected as the characteristic value. The projection length can be easily
determined based on minimum and maximum x-coordinates of the stroke and
minimum and maximum y-coordinates of the stroke.
The projection lengths for all of the strokes shown in FIG. 2 are shown in
FIG. 5. It is seen from FIG. 5 that the projection lengths of the strokes
S.sub.1 -S.sub.10 which are character strokes are short and the projection
lengths of the strokes S.sub.11 -S.sub.13 which are graphic strokes are
long. Accordingly, the projection length is compared with a predetermined
threshold and if the projection length is smaller than the threshold, it
relates to the character stroke and hence the character recognition mode
is selected, and if the comparison result is opposite, it relates to the
graphic stroke and the graphic recognition mode is selected. In the
character recognition, segmentation for segmenting the input stroke into
an appropriate number of sections is required. Accordingly, the input
means is sectioned by appropriate section as shown in FIG. 2 to effect the
segmentation by those sections. In the segmentation, the strokes in one
section are regarded as the strokes of the same character.
FIG. 6 shows a status transition of the recognition mode. Three recognition
modes null, character recognition and graphic recognition are included.
The null mode is other than the character mode and the graphic mode, and
in the null mode, the stroke information is stored in a buffer or the
stroke information is not available (recognition has not been completed).
At RESTART (1), the mode is the null mode and the input of the stroke
information is monitored. The mode is transitioned from the null mode
under condition (2), (3) or (5).
Transition under condition (2): FLM.multidot.(lm<l.sub.TH)
Transition under condition (3): FLM.multidot.(lm<l.sub.TH)
Transition under condition (5): lm.gtoreq.l.sub.TH
where lm is the projection length, l.sub.TH is the predetermined threshold,
FLM is a condition under which a start point of the stroke is present in
the same section, and FLM is the opposite condition.
In the transition condition (3), the strokes are stored in the memory
buffer until the character strokes are segmented. On the other hand, for
the graphic, the graphic and the connecting line are frequently drawn in
one stroke. Accordingly, the mode is transitioned to the graphic mode if
the characteristic value is larger than the predetermined threshold as
shown by the condition (5).
When the reoognition in the character or graphic recognition mode is
completed, the mode returns to the null mode at RESULT (4) or (6). If the
recognition result is rejected at REJECT (7) (character does not match to
characters in a dictionary), the mode is transitioned to the graphic
recognition mode and the graphic recognition is effected. This is
effective when a special graphic pattern not contained in the dictionary
is to be recognized.
FIG. 7 shows a processing flow for implementing the units 3, 4 and 5 of
FIG. 1. Since the hand-written character/graphic recognition unit 4 and
the character/graphic display unit 5 are known units, the explanation
thereof is omitted. In the following description, the mode recognition
unit 3 which is the characteristic feature of the present invention is
explained in detail.
Steps 3a-3i in FIG. 7 relate to the mode recognition unit 3. Initialization
such as clearing of tables and initialization for calculation of the
characteristic value of the stroke is first effected (step 3a), and the
hand-written strokes are inputted and stored in the buffer (step 3b). The
buffer is a ring buffer as shown on a table STROKE of FIG. 10. For
example, strokes S.sub.1 -S.sub.4 are sequentially stored. As the stroke
is inputted, an input pointer INP is updated. On the other hand, when the
recognition is completed, an output pointer OUTP is updated. The input
pointer INP cannot overrun the output pointer OUTP. If the buffer is full,
the old strokes are thrown away. Since the buffer usually has a
sufficiently large capacity (for example, 4K bytes), the stroke data is
not thrown away.
In steps 3c-3e, the characteristic value necessary for the mode recognition
is calculated based on the stroke coordinates stored in the stroke table
STROKE. A minimum coordinate P.sub.min and a maximum coordinate P.sub.max
are calculated based on the stroke coordinates (step 3c), and projection
lengths .DELTA.u and .DELTA.v on the x-axis and the y-axis are determined
(step 3d).
##EQU1##
and the larger one of the projection lengths is selected as the
characteristic value lm (step 3e).
lm=max{.DELTA.u, .DELTA.v} (2)
The characteristic value lm is compared with the predetermined threshold
l.sub.TH (step 3f), and if lm>l.sub.TH, the graphic recognition mode is
selected (step 3i). If lm<l.sub.TH, whether the start point of the stroke
is present in the same section as the previous stroke (step 3g). This
carries out the segmentation of the character. If it is present in the
same section in the step 3g, the process jumps to the step 3b where the
strokes are again inputted and stored. FIG. 10 shows a data store
condition in the stroke table when four strokes of the character " " shown
in FIG. 2 are in the same section No. 7. When the first stroke of the next
character " " is inputted, the start point of the stroke is in the next
section No. 8. Accordingly, the decision in the step 3g of FIG. 7 is NO
and the process proceeds to the step 3h where the character recognition
mode is selected.
After the mode recognition, the hand-written character/graphic recognition
is carried out (step 4). The recognized character or graphic is then
displayed on the display (step 5).
FIG. 8 shows a detailed flow chart for the minimum and maximum coordinate
calculation in the step 3c.
FIG. 9 shows a sub-flow for the hand-written character/graphic recognition
in the step 4. If the recognition result for the hand-written character is
rejected, the hand-written graphic recognition is initiated. (See the mode
transition chart of FIG. 6).
The present embodiment offers the following advantages.
First, since the characteristic value for the character/graphic mode
recognition is derived from the stroke information and it is compared with
the threshold value to set the mode, the operator need not pay attention
to the character/graphic mode when he/she input the hand-writing strokes.
According, the function switches need not be operated and the characters
and graphics can be inputted in a more natural manner and the man-machine
interface is significantly improved.
Secondly, since the mode recognition is carried out for each stroke to
recognize the character/graphic, a response is improved over the prior art
method in which the recognition is made after all of the character and
graphic have been inputted.
Thirdly, since the projection length of the stroke is used as the
characteristic value for the mode recognition, it is independent from the
complexity of the stroke and the stroke length and the character/graphic
recognition can be effectively carried out.
Fourthly, since the character stroke is segmented by the section (input
grid), the character is readily segmented.
Fifthly, since the mode is transitioned to the hand-written graphic
recognition mode if the recognition result for the hand-written character
is rejected, even a special graphic having a short stroke can be
recognized.
Further, since the combined input/display device is used, the
characters/graphics can be inputted in the same manner as if they are
inputted by a pencil and a paper and the man-machine interface is
significantly improved.
Referring to FIGS. 11 to 16, other embodiments of the present invention are
explained.
In the previous embodiment, if short bars such as strokes S.sub.3 -S.sub.6
shown in FIG. 11 are inputted, those strokes are recognized as character
strokes. In the present embodiment, when a graphic stroke S.sub.2 having a
stroke length l.sub.s is inputted, the stroke S.sub.2 is translated to a
line by the hand-written graphic recognition processing and an area As
having a width d around the translated line is defined. It is called a
graphic band. It is checked if a start point (Po) or an end point of the
stroke is present in the graphic band. If it is present, it is recognized
as a graphic stroke.
The above processing may be carried out by adding a step 3f' shown in FIG.
13 after the step 3f shown in FIG. 7. The graphic band may be generated by
adding a step 4e shown in FIG. 14 after the step 4d shown in FIG. 9.
In the present embodiment, when the short bars are inputted as described
above, they are recognized as graphic strokes.
FIG. 15 shows another embodiment. In the present embodiment, a stroke
length lm is determined as the characteristic value and it is compared
with a predetermined threshold l.sub.TH.
As shown in FIG. 16, the stroke length for the stroke S.sub.1, for example,
may be readily determined by summing distances between coordinate points
P.sub.0 -P.sub.1, P.sub.1 -P.sub.2, P.sub.2 -P.sub.3 and P.sub.3 -P.sub.4.
In the present embodiment, similar advantage to that of the previous
embodiment is offered because the character stroke is generally shorter
than the graphic stroke.
FIG. 17 shows another embodiment. Function areas 1d, 1e and 1f are provided
in the coordinate input area of the coordinate input means 1 to allow
selective designation of the character recognition mode and the graphic
recognition mode. The function area 1d is used to designate the manual
character input mode, the function area 1e is used to designate the
graphic input mode, and the function area 1f is used to switch the mode
from the selected designation to the automatic mode. If, in the automatic
mode, the character or graphic recognition mode is designated, it is
forcibly set. This is the manual mode, and the mode can be switched to the
automatic mode by depressing the function area 11f. The strokes are
inputted by the area 1c.
FIG. 18 shows a flow chart corresponding to FIG. 7, when the embodiment of
FIG. 17 is used. Difference from FIG. 7 is addition of steps 3b' and 3b'.
The present embodiment is effective when the mode designation cannot be
achieved only by the stroke coordinates, such as when a large character or
a small graphic is inputted.
* * * * *
|
|
 |
|
 |
Description |
|
