Hang Joon Kim

Automatic recognition of a car license plate using color image processing

An automatic recognition method of a car license plate using color image processing is presented. At first, background colors of a plate are extracted from an input car image. A neural network is used for more stable extraction. To find a plate region, a fixed ratio of horizontal and vertical length of a plate is used. To recognize characters in a plate, template matching and postprocessing techniques are used. Since the proposed method does not depend on line information of a plate it is very robust to boundary deformation. Also, this method is strong enough to deal with a cars image which has many similar regions with a plate.<<ETX>>

On-line recognition of handwritten chinese characters based on hidden markov models

Difficulties in Chinese character recognition due to numerous strokes usually warped into a cursive form and a much larger set of characters. In this paper, we propose a hidden Markov model (HMM) based recognition model that deals efficiently with these recognition problems. The model is an interconnection network of radical and ligature HMMs. It works well with variations of the cursive strokes by the characteristics of the HMMs. It represents the large character set with a relatively small memory and also has good extensibility. To solve the problem of recognition speed caused by a number of search paths, we combine a modified level building search with the isolated radical and ligature HMMs in an attempt to achieve a robust, accurate recognizer whose performance is optimized. The algorithm is an efficient network search procedure, the time complexity of which depends on the number of levels in the network. A test with 18,000 handwritten characters shows a recognition rate of 90.3% and a speed of 1.83 s per character.

Recognition of off-line handwritten Korean characters

Abstract In this paper, we propose an on-line model-guided recognition method for off-line handwritten Korean characters. To recognize handwritten Korean characters, it is required to extract strokes and stroke sequence to describe an input two-dimensional character as one-dimensional representation. We define 28 primitive strokes and introduce 300 stroke separation rules to ectract proper strokes from written Korean characters. To find a stroke sequence we use stroke codes and stroke relations between two consecutive strokes. The input character is recognized by searching the character recogniton trees with the stroke sequence. The proposed method has been tested with a set of the most frequently used 1000 characters written by 400 different writers and showed the recognition rate of 94%

On-line recognition of cursive Korean characters using a set of extended primitive strokes and fuzzy functions

Abstract We propose an efficient method for on-line recognition of cursive Korean characters. The method uses a set of extended primitive strokes and fuzzy membership functions to extract cursive strokes. Positional relation of two strokes is computed also using fuzzy functions. With a sequence of stroke vectors and their positional relations computed, the method searches the target character on a character database. A test with 17,500 handwritten characters shows recognition rate of 96.3% with speed of 0.3 second per character.

On-line recognition of run-on Korean characters

The authors propose an efficient method for on-line recognition of run-on Korean characters. The method uses a set of extended primitive strokes and fuzzy membership functions to extract cursive strokes. To separate a character from run-on characters they construct character separation rules and a generalized character list. With a sequence of stroke codes and their positional relations computed, the method searches the target character on a generalized character list. A test with 17500 characters shows a recognition rate of 98.4% with a speed of 0.3 second per character.

ON-LINE RECOGNITION OF CURSIVE KOREAN CHARACTERS USING ART-BASED STROKE CLASSIFICATION (RECOGNITION OF CURSIVE KOREAN CHARACTERS)

This paper proposes an efficient method for on-line recognition of cursive Korean characters. Strokes, primitive components of Korean characters, are usually warped into a cursive form and classifying them is very difficult. To deal with such cursive strokes, we consider them as a recognition unit and automatically classify them by using an ART-2 neural network. The neural network has the advantage of assembling similar patterns together to form classes in a self-organized manner. This ART-2 stroke classifier contributes to high stroke recognition rate and less recognition time. A database for character recognition is also dynamically constructed with a tree structure, and a new character can be included simply by adding a new sequence to it. Character recognition is achieved by traversing the database with a sequence of recognized strokes and positional relations between the strokes. To verify the performance of the system, we tested it on 17,500 handwritten characters, and obtained a good recognition rate of 96.8% and a speed of 0.52 second per character. This results suggest that the proposed method is pertinent to be put into practical use.

Off-line handwritten Korean character recognition based on stroke extraction and representation

Abstract In this paper, we propose an off-line recognition method for handwritten Korean characters based on stroke extraction and representation. To recognize handwritten Korean characters, it is required to extract strokes and stroke sequence to describe an input of two-dimensional character as one-dimensional representation. We define 28 primitive strokes to represent characters and introduce 300 stroke separation rules to extract proper strokes from Korean characters. To find a stroke sequence, we use stroke code and stroke relationship between consecutive strokes. The input characters are recognized by using character recognition trees. The proposed method has been tested for the most frequently used 1000 characters by 400 different writers and showed recognition rate of 94.3%.

Handwritten Korean character recognition by stroke extraction and representation

We propose an extended handwritten Korean character (Hangul) recognition algorithm by stroke extraction and representation which is based on the structure of the character and writing habits. In the case of handwritten Hangul, contact among graphemes and various abnormal graphemes are major obstacle for the useful system. To solve the problem, we define 26 primitive strokes for handwritten Hangul and build stroke separation rules. Since the vertical vowel make a lot of grapheme contacts, the system recognizes them first. After the vertical vowel recognition, stroke separation and representation grapheme recognition and character code generation steps are followed. We implement the proposed algorithm on most frequently used 2,600 characters and show the recognition result.<<ETX>>

Implementation of handwritten digit recognition system on multitransputer

We propose an off-line digit recognition algorithm through stroke classification and implement it on a multitransputer for parallel processing. We classify possible strokes in the digit into 6 strokes and make a decision tree for recognition. To enhance the performance of the recognition system the whole task is partitioned into parallel subtasks considering best trade-off between parallelism and overhead. Though parallel tasks were executed concurrently in a processor on star-networked multitransputer because of resource limits, a model of parallel recognition system is presented. By increasing the number of processors to allocate the internal parallel processes privately, parallel recognition system may be put into practical use.<<ETX>>

Hangul recognition on multi-transputer system

We propose a printed Hangul recognition algorithm on multitransputer system. The vowels in a printed Hangul have very outstanding features and are relatively easy to locate and recognize. We developed subpattern separation algorithm which is based on the vowel. After subpattern separation, the system extracts stroke codes for each subpattern and recognizes them. To get higher recognition speed, we allocated each processing step on a multitransputer system. We tested the validity of the recognition system for 7,000 printed characters from an ordinary text. It showed good result in both speed as well as recognition rates.<<ETX>>