Image analysis – Pattern recognition – Feature extraction
Reexamination Certificate
2002-09-12
2004-04-13
Mariam, Daniel (Department: 2621)
Image analysis
Pattern recognition
Feature extraction
C382S198000, C382S202000
Reexamination Certificate
active
06721452
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates generally to the field of pattern and character recognition; and more particularly to an “activity”-based system and method for feature extraction, representation and character recognition that reduces the required processing capacity for recognizing single stroke characters (or multiple strokes concatenated into one stroke) or patterns, with the intent that said characters or patterns may be created, removed, or edited from an alphabet by an individual for the purpose of personalization, without a method redesign. Further, the system and method of the present invention provide a parameter set such that its variance over an arbitrary alphabet can optimize recognition accuracy specific to that alphabet.
BACKGROUND OF THE INVENTION
Methods for character, handwriting and pattern recognition for the purpose of alphanumeric or symbolic data (collectively referred to herein as “text”) entry into computer systems has been a key research area for electrical engineers and computer scientists since the earliest days of computers. In fact, handwriting-based input systems were designed and attempted as early as about 1959, prior to the widespread use of alphanumeric keyboards. Even these systems are based on the symbol recognition technologies of about the early 1950s. Most early methods were “off-line” processing methods, which used both temporal and string contextual information to increase recognition accuracy. “On-line” recognition uses only temporal drawing information to recognize while a user is writing. Generally, on-line methods sacrifice accuracy for real-time performance speeds. That sacrifice typically is not necessary for off-line recognition.
During the bulk of the 1960s, the keyboard was the premier form of text input as well as primary human interface to the computer. With the introduction of Douglas Engelbart's “mouse” and “graphical user interface” (GUI) in 1968, and the advent of digitizing tablets in the late 1960s, focus returned to research dealing with more natural human interfaces for manipulating digitized information. This remains the trend today with the various mainstream operating systems and desktop environments such as Apple's Macintosh OS, X-Windows for the various Unix systems, and Microsoft's Windows operating systems. In these systems, the mouse or some other pointing device such as a tablet or stylus are used to visually manipulate the organization of information on a screen (e.g., moving a window from the left side of the screen to the right, or to select a block of text). The text input mechanisms to all these systems, however, is still based primarily on the keyboard.
In the modern world, computing devices are getting smaller and more powerful (sometimes exceeding the power of five year old desktop personal computers) and are cheaper to produce. These small devices require text input devices that are not as cumbersome as keyboards. One potential alternative is handwriting recognition. Devices such as Apple's Newton provided this technology, but with unacceptable performance. This is due to the complex issues of not only character recognition, but of trying to separate individual characters and symbols from handwritten words, sentences or complete documents prior to recognizing each character. Only recently has a viable solution to character separation been proposed.
In about 1993, the concept of writing characters one on top of the other in single strokes so that each character is automatically separated by “pen events” (such as pressing the pen to the writing surface to signify the start of a new character, dragging the pen along the writing surface to represent the structure of the character, and lifting the pen from the writing surface to signify the end of a character) was introduced. This reduces recognition tasks to the character level. Personal digital assistants (PDAs) like the Palm Pilot and iPaq have become mainstream and are incorporating this character recognition concept with great success.
The recognition accuracy of these devices is compromised, however, in the attempt to provide a specialized alphabet that is accessible to all users, along with a recognition method robust enough to handle the different writing styles of an arbitrary user. Palm's Graffiti language, for example forces users to learn an alphabet that is potentially different from the day-to-day alphabet they are accustomed to. This adds user error to the recognition failure rates as they may continue to draw the letter ‘Q’ as they would on paper while trying to enter text into the Palm Pilot. This is an unnecessary constraint on the user, especially those who lack the motor control required to perform some of the Graffiti strokes. This would included sufferers of Parkinson's disease, Multiple Sclerosis (MS) and Muscular Dystrophy (MD). Additionally, the Palm recognition method does not appear to be robust enough to distinguish letters like ‘U’ and ‘V’ naturally, and so a serif was added onto the tail of the ‘V’ for greater separation. While this improves the distinction between such letters, it adds even greater difficulty to learning the new alphabet. In order to avoid these unnatural characters, one recognition system adds code that, when determining that the input character was either a ‘P’ or ‘D’, compares the height of the stem to the height of the attached curve in order to properly recognize. This does improve accuracy, but suggests that additional changes to the alphabet would require more character specific code to be written to handle new similarities, thus preventing the user from updating the character dictionary herself.
Some character recognition techniques such as structural matching and elastic relaxation employ complex feature manipulation methods for converting a “sloppy” character to one that is stored in a character dictionary. These methods are difficult to comprehend and deploy by most vendors (in practice) and have high computational requirements. While the Merlin system was designed to be interpreted (Java) on weak devices such as portable phones, its incorporation of these methods detract from its speed.
Presently, most research in on-line character recognition has centered around single character entry systems. Characters are entered one at a time and the recognizer classifies the character before the next is written. This provides the user immediate feedback so that errors can be corrected as they occur. Typically, there is a simple method for the user to depict the beginning and end of each character—commonly accomplished by pen down and up events.
Unistrokes, developed at Xerox Corporation in about 1993, is a well known example of a single character, pen-event system. Unistrokes characters were designed to be written one on top of another so as to minimize the real estate required for recognition and to allow for “eyes free operation”. The Unistrokes alphabet is based on five basic strokes and their rotational deformations. While several characters (‘i’, ‘j’, ‘L’, ‘o’, ‘s’, ‘v’ and ‘z’ for example) are represented by strokes similar to their Roman drawings, most characters' strokes require memorization. Additionally, a model has been developed for predicting the time required to enter arbitrary text with Unistrokes by an expert user. This is particularly useful since several variations of the Unistrokes alphabet have been introduced over the past nine years.
Since about the mid 1990's online character recognition has become widely employed in Personal Digital Assistants (PDA's), beginning with the Palm OS device, which primarily defined the product category. A popular variation of Unistrokes is the Graffiti system used in the Palm OS family of PDA's. Graffiti improved upon Unistrokes by representing characters with symbols that are, for the most part, quite like their Roman counterparts. A disadvantage of both Graffiti and Unistrokes is that their alphabets are static. As users change applications, more or fewer characters may be required.
Chapman Richard O.
Confer William J.
Auburn University
Gardner & Groff, P.C.
Mariam Daniel
LandOfFree
System and method of handwritten character recognition does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with System and method of handwritten character recognition, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and method of handwritten character recognition will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3248766