• Journal of Computing Science and Engineering
• /
• v.12 no.1
• /
• pp.24-35
• /
• 2018

Deep convolutional neural network (DCNN) is an advanced technology in image recognition. Because of extreme computing resource requirements, DCNN implementation with software alone cannot achieve real-time requirement. Therefore, the need to implement DCNN accelerator hardware is increasing. In this paper, we present a field programmable gate array (FPGA)-based hardware accelerator design of DCNN targeting handwritten Hangul character recognition application. Also, we present design optimization techniques in SDAccel environments for searching the optimal FPGA design space. The techniques we used include memory access optimization and computing unit parallelism, and data conversion. We achieved about 11.19 ms recognition time per character with Xilinx FPGA accelerator. Our design optimization was performed with Xilinx HLS and SDAccel environment targeting Kintex XCKU115 FPGA from Xilinx. Our design outperforms CPU in terms of energy efficiency (the number of samples per unit energy) by 5.88 times, and GPGPU in terms of energy efficiency by 5 times. We expect the research results will be an alternative to GPGPU solution for real-time applications, especially in data centers or server farms where energy consumption is a critical problem.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1998.06a
• /
• pp.288-295
• /
• 1998

This paper proposes a new data classification method based on the tolerant rough set that extends the existing equivalent rough set. Similarity measure between two data is described by a distance function of all constituent attributes and they are defined to be tolerant when their similarity measure exceeds a similarity threshold value. The determination of optimal similarity theshold value is very important for the accurate classification. So, we determine it optimally by using the genetic algorithm (GA), where the goal of evolution is to balance two requirements such that (1) some tolerant objects are required to be included in the same class as many as possible. After finding the optimal similarity threshold value, a tolerant set of each object is obtained and the data set is grounded into the lower and upper approximation set depending on the coincidence of their classes. We propose a two-stage classification method that all data are classified by using the lower approxi ation at the first stage and then the non-classified data at the first stage are classified again by using the rough membership functions obtained from the upper approximation set. We apply the proposed classification method to the handwritten numeral character classification. problem and compare its classification performance and learning time with those of the feed forward neural network's back propagation algorithm.

• Journal of Korea Society of Industrial Information Systems
• /
• v.5 no.1
• /
• pp.25-30
• /
• 2000

Hangul is regarded as one of the difficult character set due to the large number of classes and the shape similarity among different characters. Most of the conventional researches attempted to recognize the 2,350 characters which are popularly used, but this approach has a problem or low recognition performance while it provides a generality. On the contrary, recognition of a small character set appearing in specific fields like postal address or bank checks is more practical approach. This paper describes a research for recognizing the handwritten Hangul characters appearing in monetary fields. The modular neural network is adopted for the classification and three kinds of feature are tested. The experiment performed using standard Hangul database PE92 showed the correct recognition rate 91.56%.

• Proceedings of the IEEK Conference
• /
• 1999.06a
• /
• pp.492-495
• /
• 1999

In this paper, a research is focused on implementation of the handwritten Korean-character recognition system using a neural coprocessor for PDA application. The proposed coprocessor is composed of a digital neural network called DMNN and a RISC-based dedicated controller in order to achieve high speed as well as compactness. Two neural networks are used for recognition, one for stroke classification out of extended 11 strokes and the other for grapheme classification. Our experimental result shows that the successful recognition rate of 92.1% over 3,000 characters written by 10 persons can be obtained. Moreover, it can be improved to 95.3% when four candidates are considered. The design verification of tile proposed neural coprocessor is conducted using the ASIC emulator for further hardware implementation.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.30B no.4
• /
• pp.1-10
• /
• 1993

In this paper, a new method of phoneme segmentation of handwritten Korean characters using the local graph pattern is proposed. At first, thinning was performed before extracting features. End-point, inflexion-point, branch-point and cross-point were extracted as features. Using these features and the angular relations between these features, local graph pattern was made. When local graph pattern is made, the of strokes is investigated on contacting point. From this process, pattern is simplified as contacting pattern of the basic form and the contacting form we must take into account can be restricted within fixed region, 4therefore phoneme segmentation not influenced by characters form and any other contact in a single character is performed as matching this local graph pattern with base patterns searched ahead. This experiments with 540 characters have been conducted. From the result of this experiment, it is shown that phoneme segmentation is independent of characters form and other contact in a single character to obtain a correct segmentation rate of 95%, manages it efficiently to reduce the time spent in lock operation when the lock.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.19 no.5
• /
• pp.883-891
• /
• 1994

In this paper, we introduce new thinning algorithm which is useful for handwritten Korean character by using pixel directivity. At first, the directivity detection is performed before thinning. Each pixel is classified into the straight line of the oblique line based on its directivity. The algorithm using Rutovitz corossing number is applied to the straight line. And the algorithm using Hilditch crossing number is applied to the oblique line. The proposed algorithm is compared with six convention algorithms. Comparison criteria are similarity, noisy branch, and phoneme segmentation rate. Experiments with 570 characters have been conducted. Experimental result shows that the proposed algorithm is superior to six conventional algorithm with respect to similarity and phoneme segmentation rate.

• Journal of The Korean Association of Information Education
• /
• v.1 no.1
• /
• pp.28-37
• /
• 1997

This paper presents a new neural network based on fuzzy set and its application to invariant character recognition. The fuzzy neural network consists of five layers. The results of simulation show that the network can recognize characters in the case of distortion, translation, rotation and different sizes of handwritten characters and even with noise(8${\sim}$30%)). Translation, distortion, different sizes and noise are achieved by layer L2 and rotation invariant by layer L5. The network can recognize 108 examples of training with 100% recognition rate when they are shifted in eight directions by 1 pixel and 2 pixels. Also, the network can recognize all the distorted characters with 100% recognition rate. The simulations show that the test patterns cover a ${\pm}20^{\circ}$ range of rotation correctly. The proposed network can also recall correctly all the learned characters with 100% recognition rate. The proposed network is simple and its learning and recall speeds are very fast. This network also works for the segmentation and recognition of handwritten characters.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.30B no.3
• /
• pp.50-59
• /
• 1993

There is basic problem with the current evaluation method for character recognition systems. The current method evaluates the average recognition rate by applying the test data to the target system. The average recognition rate tells no more than and no less than the overall performance and it depends on the data. In this paper we propose a testing method which will analyze the target system and point out its strong points and weak points. This can be made possible through using the data which are generated cy distorting the standard character images according to a carefully controlled manner. This paper will describe how to automatically generate such distorted images. Also we will show the method is actually effective and useful by applying it to evaluating existing recognition algorithms.

• Journal of Platform Technology
• /
• v.6 no.1
• /
• pp.3-8
• /
• 2018

This paper proposes a technique for recognizing online handwritten cursive data obtained by tracing a motion trajectory while a user is in the 3D space based on a convolution neural network (CNN) algorithm. There is a difficulty in recognizing the virtual character input by the user in the 3D space because it includes both the character stroke and the movement stroke. In this paper, we divide syllable into consonant and vowel units by using labeling technique in addition to the result of localizing letter stroke and movement stroke in the previous study. The coordinate information of the separated consonants and vowels are converted into image data, and Korean handwriting recognition was performed using a convolutional neural network. After learning the neural network using 1,680 syllables written by five hand writers, the accuracy is calculated by using the new hand writers who did not participate in the writing of training data. The accuracy of phoneme-based recognition is 98.9% based on convolutional neural network. The proposed method has the advantage of drastically reducing learning data compared to syllable-based learning.

• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.36C no.7
• /
• pp.27-35
• /
• 1999

Researches on the recognition of handwritten script have been conducted under the assumption that the isolated recognition units are provided as inputs. However, in practical recognition system designs, providing the isolated recognition unit is an challenge due to various writing syles. This paper proposes an approach for segmenting words from lines of unconstrained handwritten text, without help of recognition. In contrast to the conventional approaches which are based on physical gaps between connected components, clues that reflect the author's writing style, in terms of spacing, are extracted and utilized for the segmentation using a simple neural network. The clues are from character segments and include normalized heights and intervals of the segments. Effectiveness of the proposed approach compared with the conventional connected component based approaches in terms of word segmentation performance was evaluated by experiments.