• Journal of Broadcast Engineering
• /
• v.19 no.1
• /
• pp.56-63
• /
• 2014

This paper deals with a binarization for plate number recognition. The binarization process converts an image into a binary image and plays an important role for automatically recognizing plate number. The rear license plate has often a shadowed image which causes erroneous binarized image due to non-uniform illumination. In this paper, a binarization method is proposed in which the shadow line is detected in a rear plate with a shadow. And then the histogram matching is conducted for the two image separated by the shadow line. After histogram matching, two images are joined and finally Otsu method is applied for the binarization. In the experiment, the proposed algorithm shows robust performance compared to the conventional method in the presence of estimation error in the shadow line.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.28 no.12C
• /
• pp.1225-1238
• /
• 2003

We present an efficient algorithm for skew correction of business card images obtained by a PDA (personal digital assistant) camera. The proposed method is composed of four parts: block adaptive binarization (BAB), stripe generation, skew angle calculation, and image rotation. In the BAB, an input image is binarized block by block so as to lessen the effect of irregular illumination and shadow over the input image. In the stripe generation, character string clusters are generated merging adjacent characters and their strings, and then only clusters useful for skew angle calculation are output as stripes. In the skew angle calculation, the direction angles of the stripes are calculated using their central moments and then the skew angle of the input image is determined averaging the direction angles. In the image rotation, the input image is rotated by the skew angle. Experimental results shows that the proposed method yields skew correction rates of about 93% for test images of several types of business cards acquired by a PDA under various surrounding conditions.

• The Journal of the Korea Contents Association
• /
• v.11 no.6
• /
• pp.18-25
• /
• 2011

Salient object attracts more and more attention from researchers due to its important role in many fields of multimedia processing like tracking, segmentation, adaptive compression, and content-base image retrieval. Usually, a saliency map is binarized into black and white map, which is considered as the binary mask of the salient object in the image. Still, the threshold is heuristically chosen or parametrically controlled. This paper suggests using the global optimal threshold to perform saliency map thresholding. This work also considers the usage of multi-level optimal thresholds and the local adaptive thresholds in the experiments. These experimental results show that using global optimal threshold method is better than parametric controlled or local adaptive threshold method.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.17 no.7
• /
• pp.1564-1570
• /
• 2013

In this paper, we propose a robust pupil detection method using rank order filter and cross-correlation. Potential pupil candidates are detected using rank order filter. Eye region is binarized using variable threshold to find eyebrow, and pupil candidates at the eyebrow are removed. The positions of pupil candidates are corrected, the pupil candidates are grouped into pairs based on geometric constraints. A similarity measure is obtained for two eye of each pair using cross-correlation, we select a pair with the largest similarity measure as a final pupil. The experiments have been performed for 500 images of the BioID face database. The results show that it achieves the high detection rate of 96.8% and improves about 11.6% than existing method.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.14 no.5
• /
• pp.277-286
• /
• 2019

In this paper, we propose an efficient hyperplane generation technique to classify human activity from combination of events and sequence information obtained from multiple-event sensors. By generating hyperplane efficiently, our machine learning algorithm classify with less memory and run time than the LSVM (Linear Support Vector Machine) for embedded system. Because the fact that light weight and high speed algorithm is one of the most critical issue in the IoT, the study can be applied to smart home to predict human activity and provide related services. Our approach is based on reducing numbers of hyperplanes and utilizing robust string comparing algorithm. The proposed method results in reduction of memory consumption compared to the conventional ML (Machine Learning) algorithms; 252 times to LSVM and 34,033 times to LSTM (Long Short-Term Memory), although accuracy is decreased slightly. Thus our method showed outstanding performance on accuracy per hyperplane; 240 times to LSVM and 30,520 times to LSTM. The binarized image is then divided into groups, where each groups are converted to binary number, in order to reduce the number of comparison done in runtime process. The binary numbers are then converted to string. The test data is evaluated by converting to string and measuring similarity between hyperplanes using Levenshtein algorithm, which is a robust dynamic string comparing algorithm. This technique reduces runtime and enables the proposed algorithm to become 27% faster than LSVM, and 90% faster than LSTM.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.17 no.1
• /
• pp.67-75
• /
• 2024

Crosswalks play an important role for the safe movement of pedestrians in a complex urban environment. However, for the visually impaired, crosswalks can be a big risk factor. Although assistive tools such as braille blocks and acoustic traffic lights exist for safe walking, poor management can sometimes act as a hindrance to safety. This paper proposes a method to improve accuracy in a deep learning-based real-time crosswalk detection model that can be used in applications for pedestrian assistance for the disabled at the beginning. The image was binarized by utilizing the characteristic that the white line of the crosswalk image contrasts with the road surface, and through this, the crosswalk could be better recognized and the location of the crosswalk could be more accurately identified by using two models that learned the whole and the middle part of the crosswalk, respectively. In addition, it was intended to increase accuracy by creating a boundary box that recognizes crosswalks in two stages: whole and part. Through this method, additional frames that the detection model did not detect in RGB image learning from the crosswalk image could be detected.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.41 no.3
• /
• pp.73-81
• /
• 2004

To extract the creditable features in fingerprint image, many people use the thinning algorithm that has a very important position in the preprocessing. In this paper, we propose the robust parallel thinning algorithm that can preserve the connectivity of the binarized fingerprint image, make the thinnest skeleton with 1-pixel width and get near to the medial axis extremely. The proposed thinning method repeats three sub-iterations. The first sub-iteration takes off only the outer boundary pixel by using the interior points. To extract the one side skeletons, the second sub-iteration finds the skeletons with 2-pixel width. The third sub-iteration prunes the needless pixels with 2-pixel width existing in the obtained skeletons and then the proposed thinning algorithm has the robustness against the rotation and noise and can make the balanced medial axis. To evaluate the performance of the proposed thinning algorithm we compare with and analyze the previous algorithms.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.13 no.8
• /
• pp.1699-1704
• /
• 2009

If detected early enough, cervical cantor may have a good survival rate due to its preneoplastic state. However, the process is so time consuming that a medical expert can handle only a small amount of such examinations. In this paper, we propose a new nucleus extraction algorithm for uterine cervical pap smears in order to mitigate such burdens of medical experts. In the preneoplastic state cytodiagnosis images, it is important to differentiate three main areas - background, cytoplasm and nucleus. Thus, we apply lighting compensation and $3{\times}3$ mask of B channel in order to restore damaged image and remove noises respectively. The cell object is obtained from those clean binarized images with Grossfire algorithm. When there are clusters of cells, the target nucleus can be obtained with repetitive binarization of R channel brightness. In our experiment of using uterine cervical pap smears of 400 magnifications that is common in the diagnostic cytology, our method is able to extract 40 nucleus out of 45 population successfully.

• Journal of the Korean Institute of Telematics and Electronics S
• /
• v.36S no.12
• /
• pp.51-57
• /
• 1999

In this paper, an adaptive thresholding of the nonuniformly contrasted images is proposed through using the contrast pre-enhancement of the local regions and the bilinear interpolation between the local threshold values. The nonuniformly contrasted image is decomposed into 9${\times}$9 sized local regions, and the contrast is enhanced by intensifying the gray level difference of each low contrasted or blurred region. Optimal threshold values are obtained by iterative method from the gray level distribution of each contrast-enhanced local region. Discontinuities are reduced at the region of interest or at the characters by using bilinear interpolation between the neighboring threshold surfaces. Character recognition experiments are conducted using backpropagation neural network on the characters extracted from the nonuniformly contrasted document, PCB, and wafer images binarized through using the proposed thresholding and the conventional thresholding methods, and the results prove the relative effectiveness of the proposed scheme.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2009.01a
• /
• pp.547-550
• /
• 2009

The object recognition mechanism of human being is not well understood yet. On research of animal experiment using an ape, however, neurons that respond to simple shape (e.g. circle, triangle, square and so on) were found. And Hypothesis has been set up as human being may recognize object as combination of such simple shapes. That mechanism is called Figure Alphabet Hypothesis, and those simple shapes are called Figure Alphabet. As one way to research object recognition algorithm, we focused attention to this Figure Alphabet Hypothesis. Getting idea from it, we proposed the feature extraction algorithm for object recognition. In this paper, we described recognition of binarized images of multifont alphabet characters by the recognition model which combined three-layered neural network in the feature extraction algorithm. First of all, we calculated the difference between the learning image data set and the template by the feature extraction algorithm. The computed finite difference is a feature quantity of the feature extraction algorithm. We had it input the feature quantity to the neural network model and learn by backpropagation (BP method). We had the recognition model recognize the unknown image data set and found the correct answer rate. To estimate the performance of the contriving recognition model, we had the unknown image data set recognized by a conventional neural network. As a result, the contriving recognition model showed a higher correct answer rate than a conventional neural network model. Therefore the validity of the contriving recognition model could be proved. We'll plan the research a recognition of natural image by the contriving recognition model in the future.