• Journal of the Korea Society of Computer and Information
• /
• v.11 no.3
• /
• pp.73-78
• /
• 2006

Conventional caption extraction methods use the difference between frames or color segmentation methods from the whole image. Because these methods depend heavily on heuristics, we should have a priori knowledge of the captions to be extracted. Also they are difficult to implement. In this paper, we propose a method that uses little heuristic and simplified algorithm. We use topographical features of characters to extract the character points and use MST(Minimum Spanning Tree) to extract the candidate regions for captions. Character regions are determined by testing several conditions and verifying those candidate regions. Experimental results show that the candidate region extraction rate is 100%, and the character region extraction rate is 98.2%. And then we can see the results that caption area in complex images is well extracted.

• Journal of Intelligence and Information Systems
• /
• v.26 no.2
• /
• pp.1-25
• /
• 2020

In this paper, we suggest an application system architecture which provides accurate, fast and efficient automatic gasometer reading function. The system captures gasometer image using mobile device camera, transmits the image to a cloud server on top of private LTE network, and analyzes the image to extract character information of device ID and gas usage amount by selective optical character recognition based on deep learning technology. In general, there are many types of character in an image and optical character recognition technology extracts all character information in an image. But some applications need to ignore non-of-interest types of character and only have to focus on some specific types of characters. For an example of the application, automatic gasometer reading system only need to extract device ID and gas usage amount character information from gasometer images to send bill to users. Non-of-interest character strings, such as device type, manufacturer, manufacturing date, specification and etc., are not valuable information to the application. Thus, the application have to analyze point of interest region and specific types of characters to extract valuable information only. We adopted CNN (Convolutional Neural Network) based object detection and CRNN (Convolutional Recurrent Neural Network) technology for selective optical character recognition which only analyze point of interest region for selective character information extraction. We build up 3 neural networks for the application system. The first is a convolutional neural network which detects point of interest region of gas usage amount and device ID information character strings, the second is another convolutional neural network which transforms spatial information of point of interest region to spatial sequential feature vectors, and the third is bi-directional long short term memory network which converts spatial sequential information to character strings using time-series analysis mapping from feature vectors to character strings. In this research, point of interest character strings are device ID and gas usage amount. Device ID consists of 12 arabic character strings and gas usage amount consists of 4 ~ 5 arabic character strings. All system components are implemented in Amazon Web Service Cloud with Intel Zeon E5-2686 v4 CPU and NVidia TESLA V100 GPU. The system architecture adopts master-lave processing structure for efficient and fast parallel processing coping with about 700,000 requests per day. Mobile device captures gasometer image and transmits to master process in AWS cloud. Master process runs on Intel Zeon CPU and pushes reading request from mobile device to an input queue with FIFO (First In First Out) structure. Slave process consists of 3 types of deep neural networks which conduct character recognition process and runs on NVidia GPU module. Slave process is always polling the input queue to get recognition request. If there are some requests from master process in the input queue, slave process converts the image in the input queue to device ID character string, gas usage amount character string and position information of the strings, returns the information to output queue, and switch to idle mode to poll the input queue. Master process gets final information form the output queue and delivers the information to the mobile device. We used total 27,120 gasometer images for training, validation and testing of 3 types of deep neural network. 22,985 images were used for training and validation, 4,135 images were used for testing. We randomly splitted 22,985 images with 8:2 ratio for training and validation respectively for each training epoch. 4,135 test image were categorized into 5 types (Normal, noise, reflex, scale and slant). Normal data is clean image data, noise means image with noise signal, relfex means image with light reflection in gasometer region, scale means images with small object size due to long-distance capturing and slant means images which is not horizontally flat. Final character string recognition accuracies for device ID and gas usage amount of normal data are 0.960 and 0.864 respectively.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2001.05a
• /
• pp.625-628
• /
• 2001

In this paper, it is shown that the car number plate are recognized more efficiently by using Fuzzy-ARTM AP. We use the location information of characters in the car number plate area and the color intensity difference between the character region and the background region int the tar number plate area. For segmented plate region, the car plate region is extracted by deciding the X-axis region composed by horizontal histogram and the Y-axis region composed by the variance histogram of vertical histogram. Our method then directly recognizes the extracted character region by using Fuzzy-ARTMAP neural network.

• Proceedings of the IEEK Conference
• /
• 2000.07b
• /
• pp.906-909
• /
• 2000

Conventional caption extraction methods use the difference between frames or color segmentation methods from the whole image. Because these methods depend heavily on heuristics, we should have a priori knowledge of the captions to be extracted. Also they are difficult to implement. In this paper, we propose a method that uses little heuristics and simplified algorithm. We use topographical features of characters to extract the character points and use KMST(Kruskal minimum spanning tree) to extract the candidate regions for captions. Character regions are determined by testing several conditions and verifying those candidate regions. Experimental results show that the candidate region extraction rate is 100%, and the character region extraction rate is 98.2%. And then we can see the results that caption area in complex images is well extracted.

• The Transactions of the Korea Information Processing Society
• /
• v.7 no.10
• /
• pp.3155-3163
• /
• 2000

In the recognitionsystem of car license plate, it is very important to extract the character from the license plate and recognize the extrated character. In this paper, I use the adaptive algorithm to recognize the charactor of licensse plate image. The adaptive algorithm is compounded of thinning algorithm template matching,algarthm, vector algorithm and so on. The adaptive algorithm was used to recognize the character from license image. In the result of expenment, character recognition is about up to 90% with the adaptive algorithm for the character region.

• Journal of Intelligence and Information Systems
• /
• v.8 no.1
• /
• pp.15-25
• /
• 2002

This paper proposes a vehicle license plate recognition system, which has relatively a simple structure and is highly tolerant of noise, by using the DCT(Discrete Cosine Transform) coefficients extracted from the character region of a license plate and the LVQ(Learning Vector Quantization) neural network. The image of a license plate is taken from a captured vehicle image based on RGB color information, and the character region is derived by the histogram of the license plate and the relative position of individual characters in the plate. The feature vector obtained by the DCT of extracted character region is utilized as an input to the LVQ neural classifier fur the recognition process. In the experiment, 109 vehicle images captured under various types of circumstances were tested with the proposed method, and the relatively high extraction rate of license plates and recognition rate were achieved.

• Proceedings of the KIEE Conference
• /
• 1987.07b
• /
• pp.936-940
• /
• 1987

This paper describes the recognition boundary of Hangeul character to interpret variance factor in accordance with various combination. When the recognition algorithm based on comparing the MSE value with the one of the standard pattern in $16{\times}16$ images is performed, we come to a conclusion that we have towe must make a decision MSE value above 34 in order to achive theve recognition rate larger than 90%. Also we understand that varing component coordinates method based on statistical process of each character pattern is preferred.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2012.04a
• /
• pp.429-431
• /
• 2012

Vehicle-logo recognition technology is very important in vehicle automatic recognition technique. The intended application is automatic recognition of vehicle type for secure access and traffic monitoring applications, a problem not hitherto considered at such a level of accuracy. Vehicle-logo recognition can improve Vehicle type recognition accuracy. So in this paper, introduces how to vehicle-logo recognition. First introduces the region of the license plate by algorithm and roughly located the region of car emblem based on the relationship of license plate and car emblem. Then located the car emblem with precision by the distance of Hausdorff. On the base, processing the region by morphologic, edge detection, analysis of connectivity and pick up the PCA character by lowing the dimension of the image and unifying the PCA character. At last the logo can be recognized using the algorithm of support vector machine. Experimental results show the effectiveness of the proposed method.

• Journal of the Korea Convergence Society
• /
• v.8 no.8
• /
• pp.33-40
• /
• 2017

In this paper, we propose a text extraction method by pattern analysis on contour for effective text detection in image. Text extraction algorithms using edge based methods show good performance in images with simple backgrounds, The images of complex background has a poor performance shortcomings. The proposed method simplifies the color of the image by using K-means clustering in the preprocessing process to detect the character region in the image. Enhance the boundaries of the object through the High pass filter to improve the inaccuracy of the boundary of the object in the color simplification process. Then, by using the difference between the expansion and erosion of the morphology technique, the edges of the object is detected, and the character candidate region is discriminated by analyzing the pattern of the contour portion of the acquired region to remove the unnecessary region (picture, background). As a final result, we have shown that the characters included in the candidate character region are extracted by removing unnecessary regions.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.10 no.4
• /
• pp.680-686
• /
• 2006

Document images captured by scanners have only skewing distortion. But camera captured document images have not only skew but also vignetting effect and geometric distortion. Vignetting effect, which makes the border areas to be darker than the center of the image, make it difficult to separate characters from the document images. But this effect has being decreased, as the lens manufacturing skill is developed. Geometric distortion, occurred by the mismatch of angle and center position between the document image and the camera, make the shape of characters to be distorted, so that the character recognition is more difficult than the case of using scanner. In this paper, we propose a method that can increase the performance of character recognition by correcting the geometric distortion of document images using a linear approximation which changes the quadrilateral region to the rectangle one. The proposed method also determine the quadrilateral transform region automatically, using the alignment of character lines and the skewed angles of characters located in the edges of each character line. Proposed method, therefore, can correct the geometric distortion without getting positional information from camera.