Pub Date : 2015-05-11DOI: 10.1109/FCV.2015.7103713
Kenji Terabayashi, T. Oiwa, K. Umeda
This paper reports that uncertainty of detected edges on a fish-eye image depends on the direction to observe the edges. In fish-eye cameras, dimensions of observation space corresponding to a pixel greatly change with the pixel location. This dimension is defined as “spatial uncertainty” in this paper and formulated with typical projection models of fish-eye cameras. Experimental results showed that the uncertainty of edge detection on fish-eye images increased with increase of the spatial uncertainty.
{"title":"Variation of edge detection uncertainty on fish-eye images","authors":"Kenji Terabayashi, T. Oiwa, K. Umeda","doi":"10.1109/FCV.2015.7103713","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103713","url":null,"abstract":"This paper reports that uncertainty of detected edges on a fish-eye image depends on the direction to observe the edges. In fish-eye cameras, dimensions of observation space corresponding to a pixel greatly change with the pixel location. This dimension is defined as “spatial uncertainty” in this paper and formulated with typical projection models of fish-eye cameras. Experimental results showed that the uncertainty of edge detection on fish-eye images increased with increase of the spatial uncertainty.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"27 15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126311602","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2015-05-11DOI: 10.1109/FCV.2015.7103722
Jung-Jae Yu, S. Han
In this paper, a novel method to automatically detect mosaic masking regions in an input image is proposed. Mosaic masking region can be used as an important clue for recognizing commercial pornographic images. The proposed method is composed of three steps. The first step is to extract SRE features using a new cross-shaped feature characteristic, and the second step is to estimate the parameters of a mosaic candidate region, and the final step is SRD verification using the characteristic of luminance distribution in a mosaic masking region. Proposed method is fast and robust to blurring effects caused by image resizing and low quality video compression and it can be used for computer vision application to block adult videos.
{"title":"Robust detection of mosaic masking region","authors":"Jung-Jae Yu, S. Han","doi":"10.1109/FCV.2015.7103722","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103722","url":null,"abstract":"In this paper, a novel method to automatically detect mosaic masking regions in an input image is proposed. Mosaic masking region can be used as an important clue for recognizing commercial pornographic images. The proposed method is composed of three steps. The first step is to extract SRE features using a new cross-shaped feature characteristic, and the second step is to estimate the parameters of a mosaic candidate region, and the final step is SRD verification using the characteristic of luminance distribution in a mosaic masking region. Proposed method is fast and robust to blurring effects caused by image resizing and low quality video compression and it can be used for computer vision application to block adult videos.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115459498","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2015-05-11DOI: 10.1109/FCV.2015.7103734
Ayaka Kojima, T. Ozeki
There are a large number of colors to represent images (e.g. 256256256 = 16,777,216 colors in an RGB color space) on computers. Since there are too many colors to handle, a large number of colors are reduced by quantization in the image processing in general. When we perform a uniform color quantization, we often get colors which do not fit the real world. Therefore, typical colors should be learned from real world images to generate a practical color palette. The bag-of-visual words based only on local features of grayscale pixel values provides the state of the art technology in the field of image classification, retrieval and recognition. Therefore it is expected to improve the performance by adding the color information to the local features. However, if we increase the number of features to extract from images, it costs memory and time for computation. Moreover, the increase of features affects the performance of recognition. The aim of this paper is to generate appropriate color palette for image classification by the bag-of-colors with less computation time and as few colors as possible to improve the accuracy of image classification.
{"title":"Color palette generation for image classification by bag-of-colors","authors":"Ayaka Kojima, T. Ozeki","doi":"10.1109/FCV.2015.7103734","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103734","url":null,"abstract":"There are a large number of colors to represent images (e.g. 256256256 = 16,777,216 colors in an RGB color space) on computers. Since there are too many colors to handle, a large number of colors are reduced by quantization in the image processing in general. When we perform a uniform color quantization, we often get colors which do not fit the real world. Therefore, typical colors should be learned from real world images to generate a practical color palette. The bag-of-visual words based only on local features of grayscale pixel values provides the state of the art technology in the field of image classification, retrieval and recognition. Therefore it is expected to improve the performance by adding the color information to the local features. However, if we increase the number of features to extract from images, it costs memory and time for computation. Moreover, the increase of features affects the performance of recognition. The aim of this paper is to generate appropriate color palette for image classification by the bag-of-colors with less computation time and as few colors as possible to improve the accuracy of image classification.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125926050","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2015-05-11DOI: 10.1109/FCV.2015.7103701
Jung Hyun, Hae-Kwang Kim, Weon Gun Oh
This paper presents results on the study of MPEG-7 visual descriptors for deformable object retrieval. A database of 819 handbag images with shape masks are constructed with different variations such as morphing, illumination changes, view point changes and color changes. For color descriptors, all of 4 MPEG-7 color descriptors of Dominant Color descriptor, Color Structure descriptor, Color Layout descriptor and Scalable Color descriptor are tested. For texture descriptor, Homogeneous Texture descriptor and Edge Histogram descriptor are tested. For shape descriptors, Contour-based and Region based descriptor are tested. The retrieval rate of the descriptors and correlation of each pair of descriptors are studied. The result shows that the Scalable Color descriptor is the best in terms of retrieval rate and the color descriptors are relatively highly correlated among themselves.
{"title":"Study on performance of MPEG-7 visual descriptors for deformable object retrieval","authors":"Jung Hyun, Hae-Kwang Kim, Weon Gun Oh","doi":"10.1109/FCV.2015.7103701","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103701","url":null,"abstract":"This paper presents results on the study of MPEG-7 visual descriptors for deformable object retrieval. A database of 819 handbag images with shape masks are constructed with different variations such as morphing, illumination changes, view point changes and color changes. For color descriptors, all of 4 MPEG-7 color descriptors of Dominant Color descriptor, Color Structure descriptor, Color Layout descriptor and Scalable Color descriptor are tested. For texture descriptor, Homogeneous Texture descriptor and Edge Histogram descriptor are tested. For shape descriptors, Contour-based and Region based descriptor are tested. The retrieval rate of the descriptors and correlation of each pair of descriptors are studied. The result shows that the Scalable Color descriptor is the best in terms of retrieval rate and the color descriptors are relatively highly correlated among themselves.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"42 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114256740","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2015-05-11DOI: 10.1109/FCV.2015.7103729
Heechul Jung, Sihaeng Lee, Sunjeong Park, Byungju Kim, Junmo Kim, Injae Lee, C. Ahn
Deep learning is considered to be a breakthrough in the field of computer vision, since most of the world records of the recognition tasks are being broken. In this paper, we try to apply such deep learning techniques to recognizing facial expressions that represent human emotions. The procedure of our facial expression recognition system is as follows: First, face is detected from input image using Haar-like features. Second, the deep network is used for recognizing facial expression using detected faces. In this step, two different deep networks can be used such as deep neural network and convolutional neural network. Consequently, we compared experimentally two types of deep networks, and the convolutional neural network had better performance than deep neural network.
{"title":"Development of deep learning-based facial expression recognition system","authors":"Heechul Jung, Sihaeng Lee, Sunjeong Park, Byungju Kim, Junmo Kim, Injae Lee, C. Ahn","doi":"10.1109/FCV.2015.7103729","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103729","url":null,"abstract":"Deep learning is considered to be a breakthrough in the field of computer vision, since most of the world records of the recognition tasks are being broken. In this paper, we try to apply such deep learning techniques to recognizing facial expressions that represent human emotions. The procedure of our facial expression recognition system is as follows: First, face is detected from input image using Haar-like features. Second, the deep network is used for recognizing facial expression using detected faces. In this step, two different deep networks can be used such as deep neural network and convolutional neural network. Consequently, we compared experimentally two types of deep networks, and the convolutional neural network had better performance than deep neural network.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"16 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114315077","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2015-05-11DOI: 10.1109/FCV.2015.7103698
Y. Iguchi, J. Yamaguchi
In this paper, the authors describe on an omnidirectional 3D measurement using double fish-eye stereo vision. This vision system consists of two stereo visions which are back to back each other. In the method, the image from the fish-eye vision is transformed into panoramic image and a parallax is detected by simple template matching to the transformed image. Using the method, three dimensional measurement of the all-directional space is possible, though the system composition is simple. In the paper, the method is explained and an example of experiment is shown.
{"title":"Omni-directional 3D measurement using double fish-eye stereo vision","authors":"Y. Iguchi, J. Yamaguchi","doi":"10.1109/FCV.2015.7103698","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103698","url":null,"abstract":"In this paper, the authors describe on an omnidirectional 3D measurement using double fish-eye stereo vision. This vision system consists of two stereo visions which are back to back each other. In the method, the image from the fish-eye vision is transformed into panoramic image and a parallax is detected by simple template matching to the transformed image. Using the method, three dimensional measurement of the all-directional space is possible, though the system composition is simple. In the paper, the method is explained and an example of experiment is shown.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130869784","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2015-05-11DOI: 10.1109/FCV.2015.7103743
Jung Hyun, Hae-Kwang Kim, Weon Gun Oh
This paper presents a new way of text region detection on the basis of Canny edge detection and connected component. A Canny edge image is detected from a gray image obtained from an original color image. The Canny image is partitioned into n × n blocks and each n × n block is divided into smaller m × m blocks. If there are sufficient edge pixels in the m × m block, then the block is set to text candidate block. The number of text candidate blocks is counted in each n × n block, and the number is sufficient, then the n × n block is set to candidate text n × n block. Text regions are only detected in the candidate text n × n blocks. Connected-component is obtained from the edge pixels in the candidate text n × n blocks of the Canny edge image. The connected components are sorted with its size and grouped in to several groups. From each group, possible candidate text lines of connected components are detected and the connected components in the neighboring groups are added into the candidate text lines. The performance of proposed method is compared with the SWT (Stroke Width Transform) and Tesseraci text region detection method. The experimental results show that proposed one is faster than SWT losing accuracy and is slower than Tesseraci with better precision.
{"title":"Fast text line detection by finding linear connected components on Canny edge image","authors":"Jung Hyun, Hae-Kwang Kim, Weon Gun Oh","doi":"10.1109/FCV.2015.7103743","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103743","url":null,"abstract":"This paper presents a new way of text region detection on the basis of Canny edge detection and connected component. A Canny edge image is detected from a gray image obtained from an original color image. The Canny image is partitioned into n × n blocks and each n × n block is divided into smaller m × m blocks. If there are sufficient edge pixels in the m × m block, then the block is set to text candidate block. The number of text candidate blocks is counted in each n × n block, and the number is sufficient, then the n × n block is set to candidate text n × n block. Text regions are only detected in the candidate text n × n blocks. Connected-component is obtained from the edge pixels in the candidate text n × n blocks of the Canny edge image. The connected components are sorted with its size and grouped in to several groups. From each group, possible candidate text lines of connected components are detected and the connected components in the neighboring groups are added into the candidate text lines. The performance of proposed method is compared with the SWT (Stroke Width Transform) and Tesseraci text region detection method. The experimental results show that proposed one is faster than SWT losing accuracy and is slower than Tesseraci with better precision.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"24 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122154206","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2015-05-11DOI: 10.1109/FCV.2015.7103716
Ryuhei Noguchi, J. Hayashi
In order to obtain the portability of a book, and reduction of a storage place, it is popular to change a book into a digital book. It has changed into the text data for search of the contents, and data size compression. However, the special character is used for the illustration page or the cover. These characters may be unable to be read as text data by the conventional character recognition. In this study, we focused on that the gradient of brightness value was flat in character area. How to presume a character area from the amount of change of the projection acquired by changing the threshold value of binarization dynamically is described.
{"title":"A method for character and photograph segmentation using dynamic thresholding","authors":"Ryuhei Noguchi, J. Hayashi","doi":"10.1109/FCV.2015.7103716","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103716","url":null,"abstract":"In order to obtain the portability of a book, and reduction of a storage place, it is popular to change a book into a digital book. It has changed into the text data for search of the contents, and data size compression. However, the special character is used for the illustration page or the cover. These characters may be unable to be read as text data by the conventional character recognition. In this study, we focused on that the gradient of brightness value was flat in character area. How to presume a character area from the amount of change of the projection acquired by changing the threshold value of binarization dynamically is described.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123142957","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2015-05-11DOI: 10.1109/FCV.2015.7103733
Seungin Baek, Soowoong Jeong, Jongsoo Choi, Sangkeun Lee
Switching median filter is known as one of effective algorithms for impulse noise reduction. In this paper, we present an improved switching median filter by considering weighted neighboring pixel locations. Specifically, the proposed method generates a flag map using boundary discriminative noise detection(BDND) detector. Next, we conduct a noise reduction by estimating the local noise density. When the local noise density is low, a corrupted pixel is replaced with the median value of uncorrupted neighboring pixels. In contrast, when the density is high, a noise searching window size increases until the predefined conditions are met. Then, a noise pixel is corrected by the weighted average of the uncorrupted values. Experiment results show that the proposed method outperforms the existing methods by about 0.5-3.7 dB on average.
{"title":"Impulse noise reduction using distance weighted average filter","authors":"Seungin Baek, Soowoong Jeong, Jongsoo Choi, Sangkeun Lee","doi":"10.1109/FCV.2015.7103733","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103733","url":null,"abstract":"Switching median filter is known as one of effective algorithms for impulse noise reduction. In this paper, we present an improved switching median filter by considering weighted neighboring pixel locations. Specifically, the proposed method generates a flag map using boundary discriminative noise detection(BDND) detector. Next, we conduct a noise reduction by estimating the local noise density. When the local noise density is low, a corrupted pixel is replaced with the median value of uncorrupted neighboring pixels. In contrast, when the density is high, a noise searching window size increases until the predefined conditions are met. Then, a noise pixel is corrected by the weighted average of the uncorrupted values. Experiment results show that the proposed method outperforms the existing methods by about 0.5-3.7 dB on average.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"26 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130972939","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2015-05-11DOI: 10.1109/FCV.2015.7103719
A. Filonenko, Danilo Cáceres Hernández, K. Jo
This paper describes the smoke detection for static cameras. The background subtraction was used to determine moving objects. Color characteristics were utilized to distinguish smoke regions and other scene members. Separate pixels were united into blobs by morphology operations and connected components labeling methods. The image is then refined by boundary roughness and edge density to decrease amount of false detections. Results of the current frame are compared to the previous one in order to check the behavior of objects in time domain.
{"title":"Smoke detection for static cameras","authors":"A. Filonenko, Danilo Cáceres Hernández, K. Jo","doi":"10.1109/FCV.2015.7103719","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103719","url":null,"abstract":"This paper describes the smoke detection for static cameras. The background subtraction was used to determine moving objects. Color characteristics were utilized to distinguish smoke regions and other scene members. Separate pixels were united into blobs by morphology operations and connected components labeling methods. The image is then refined by boundary roughness and edge density to decrease amount of false detections. Results of the current frame are compared to the previous one in order to check the behavior of objects in time domain.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"10 11","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133008965","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：