Pub Date : 2015-05-11DOI: 10.1109/FCV.2015.7103706
Tomoya Kondo, K. Kato, Kazuhiko Yamamoto
Many skin region detection methods have been proposed, and color information of the skin is an easy method to detect the skin region from the image. However, the color of the skin is not stable, because the illumination condition, especially the ambient light changes often. Usually, the color space is used to represent the skin color. Nevertheless, it is needed to adjust the parameters of the skin color in the color space for the ambient light change. In order to solve this problem, we propose two methods to detect skin region by estimating the color of the ambient light. In our method, the color of the ambient light is estimated by using one reference light. In the first method, the color of the ambient light is estimated by subtracting two images; an image taken under the unknown ambient light, and an image irradiated a reference light. To subtract these two images, the influence of the ambient light is suppressed. Then the color of the ambient light is estimated. However, irradiation of the reference light is required by changing the ambient light in this method. In the second method, the color of the ambient light is estimated by using the color of the object in the image. However, usually, there is not an object whose color is known. Thus the color of the object in the image is estimated by irradiating the reference light, first. If there is an immovable object in the image, for example the background, this method can estimate the color of the ambient light without irradiate the reference light again. In this paper, these two methods are described, and comparative experiments were conducted. By experimental results, we confirmed that the skin region was detected correctly under the several ambient light environments.
{"title":"A proposal of ambient light estimation methods for skin region detection","authors":"Tomoya Kondo, K. Kato, Kazuhiko Yamamoto","doi":"10.1109/FCV.2015.7103706","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103706","url":null,"abstract":"Many skin region detection methods have been proposed, and color information of the skin is an easy method to detect the skin region from the image. However, the color of the skin is not stable, because the illumination condition, especially the ambient light changes often. Usually, the color space is used to represent the skin color. Nevertheless, it is needed to adjust the parameters of the skin color in the color space for the ambient light change. In order to solve this problem, we propose two methods to detect skin region by estimating the color of the ambient light. In our method, the color of the ambient light is estimated by using one reference light. In the first method, the color of the ambient light is estimated by subtracting two images; an image taken under the unknown ambient light, and an image irradiated a reference light. To subtract these two images, the influence of the ambient light is suppressed. Then the color of the ambient light is estimated. However, irradiation of the reference light is required by changing the ambient light in this method. In the second method, the color of the ambient light is estimated by using the color of the object in the image. However, usually, there is not an object whose color is known. Thus the color of the object in the image is estimated by irradiating the reference light, first. If there is an immovable object in the image, for example the background, this method can estimate the color of the ambient light without irradiate the reference light again. In this paper, these two methods are described, and comparative experiments were conducted. By experimental results, we confirmed that the skin region was detected correctly under the several ambient light environments.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"5 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":"116926092","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 : 2015-05-07DOI: 10.1109/FCV.2015.7103717
Y. Minami, K. Enomoto, M. Migita, M. Toda
In this paper, we present a vivid image from a foggy underwater image. On fisheries research, it is important to grasp the situation of underwater. They do the undersea observation by using the underwater camera. However, dust floating underwater or mud alter the scene appearance of the camera. They scatter incident light on the camera. The light scattered is one of the major factor of degradation of the scene quality. We tried to solve this problem by adopting defogging algorithm to the low-quality underwater images. But in the underwater images, the environment is very different from the images on ground. The images on ground, airlight is distributed uniformly. On the other hand, the images in underwater, airlight is significantly different between each frame or partial region. We propose the algorithm that can improve visibility from a underwater image which have violent fluctuations of airlight.
{"title":"Spatially adaptive image defogging using color characteristics","authors":"Y. Minami, K. Enomoto, M. Migita, M. Toda","doi":"10.1109/FCV.2015.7103717","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103717","url":null,"abstract":"In this paper, we present a vivid image from a foggy underwater image. On fisheries research, it is important to grasp the situation of underwater. They do the undersea observation by using the underwater camera. However, dust floating underwater or mud alter the scene appearance of the camera. They scatter incident light on the camera. The light scattered is one of the major factor of degradation of the scene quality. We tried to solve this problem by adopting defogging algorithm to the low-quality underwater images. But in the underwater images, the environment is very different from the images on ground. The images on ground, airlight is distributed uniformly. On the other hand, the images in underwater, airlight is significantly different between each frame or partial region. We propose the algorithm that can improve visibility from a underwater image which have violent fluctuations of airlight.","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":"2015-05-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128847552","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 : 2015-05-07DOI: 10.1109/FCV.2015.7103754
I. Wijaya, I. B. K. Widiartha, K. Uchimura, G. Koutaki
The paper presents a pornographic image recognition using fusion of scale invariant descriptor. The pornographic image means the image contains and shows genital elements of human body having large variability due to poses, lighting, and backgrounds variations. The fusion of scale invariant descriptor that is holistic feature is employed to handle those variability problems. This holistic feature that is pose and scale invariant information of pornographic images is extracted by fusing the scale invariant descriptor of skin region of interests (ROIs) of pornographic images. The skin ROI is used to handle the large variability of pornographic images due to background variations. The main aim of this research finds a good solution for pornographic recognition system, which can be developed to limit the accessing pornographic images by teenagers and children. The experimental results show that the proposed method tends to provide high enough accuracy more than 80%, small enough FNR and FPR bout 2.77% and 28.79%, respectively. It means the proposed method is suitable to develop rejection system of pornographic images. Furthermore, these achievements are much better than the achievements of established methods. This results can be achieved because the fusion of scale invariant descriptor consists rich pornographic information representing holistic feature of pornographic images.
{"title":"Phonographic image recognition using fusion of scale invariant descriptor","authors":"I. Wijaya, I. B. K. Widiartha, K. Uchimura, G. Koutaki","doi":"10.1109/FCV.2015.7103754","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103754","url":null,"abstract":"The paper presents a pornographic image recognition using fusion of scale invariant descriptor. The pornographic image means the image contains and shows genital elements of human body having large variability due to poses, lighting, and backgrounds variations. The fusion of scale invariant descriptor that is holistic feature is employed to handle those variability problems. This holistic feature that is pose and scale invariant information of pornographic images is extracted by fusing the scale invariant descriptor of skin region of interests (ROIs) of pornographic images. The skin ROI is used to handle the large variability of pornographic images due to background variations. The main aim of this research finds a good solution for pornographic recognition system, which can be developed to limit the accessing pornographic images by teenagers and children. The experimental results show that the proposed method tends to provide high enough accuracy more than 80%, small enough FNR and FPR bout 2.77% and 28.79%, respectively. It means the proposed method is suitable to develop rejection system of pornographic images. Furthermore, these achievements are much better than the achievements of established methods. This results can be achieved because the fusion of scale invariant descriptor consists rich pornographic information representing holistic feature of pornographic images.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-05-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125203887","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 : 1900-01-01DOI: 10.1109/FCV.2015.7103709
Hayato Iwasa, W. Ohyama, T. Wakabayashi, F. Kimura
Extraction of facial feature points is essential task for many kinds of applications of face images. The feasibility of facial features extraction is determined by not only its accuracy but processing time. Some applications require real-time detection of facial features. This research aims to propose a method of facial features extraction by an accelerated implementation of circular Hough transform with gradients and appearance evaluation by histogram of gradient features. The acceleration implementation employs General Purpose computing on Graphics Processing Unit. Experiment using FERET database shows that the proposed method successfully extracted eyes and nose for 98.44% and 99.50% of frontal face images in the dataset. And 96.5% of computational time was reduced by accelerated implementation employing GPGPU.
{"title":"Facial features extraction by accelerated implementation of circular hough transform and appearance evaluation","authors":"Hayato Iwasa, W. Ohyama, T. Wakabayashi, F. Kimura","doi":"10.1109/FCV.2015.7103709","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103709","url":null,"abstract":"Extraction of facial feature points is essential task for many kinds of applications of face images. The feasibility of facial features extraction is determined by not only its accuracy but processing time. Some applications require real-time detection of facial features. This research aims to propose a method of facial features extraction by an accelerated implementation of circular Hough transform with gradients and appearance evaluation by histogram of gradient features. The acceleration implementation employs General Purpose computing on Graphics Processing Unit. Experiment using FERET database shows that the proposed method successfully extracted eyes and nose for 98.44% and 99.50% of frontal face images in the dataset. And 96.5% of computational time was reduced by accelerated implementation employing GPGPU.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"38 2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126267260","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 : 1900-01-01DOI: 10.1109/FCV.2015.7103749
T. Funahashi, K. Taki, H. Koshimizu, Akito Kaneko
In this paper, we propose a new visual inspection method that can analyze the exterior surface of rubber tires with the light stripe projection. Image sensing of the tire surface was implemented by setting the tire on the rotating table so that the normal pattern of the tire surface could be suppressed. The proposed method detects the exterior thin defect after removing three kinds of uneven circularities due to the real shape of tire, the eccentricity of rotating table and the inconsistency between two. Thus, in this research, we realized a visual inspection method by transforming three-dimensional shape of tire surface into two-dimensional image information through the analysis of light stripe images.
{"title":"Fast and robust visual inspection system for tire surface thin defect","authors":"T. Funahashi, K. Taki, H. Koshimizu, Akito Kaneko","doi":"10.1109/FCV.2015.7103749","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103749","url":null,"abstract":"In this paper, we propose a new visual inspection method that can analyze the exterior surface of rubber tires with the light stripe projection. Image sensing of the tire surface was implemented by setting the tire on the rotating table so that the normal pattern of the tire surface could be suppressed. The proposed method detects the exterior thin defect after removing three kinds of uneven circularities due to the real shape of tire, the eccentricity of rotating table and the inconsistency between two. Thus, in this research, we realized a visual inspection method by transforming three-dimensional shape of tire surface into two-dimensional image information through the analysis of light stripe images.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134584388","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 : 1900-01-01DOI: 10.1109/FCV.2015.7103732
Shubhra Aich, Yong-Cheol Lee, Chil-Woo Lee
In this paper, we propose a supervised approach to find out the probabilistic mapping of semantic contours in color images. We prepare a new color image modifying the RGB color planes to incorporate reasonable within-object contrasts in all the color planes. Color gradient based features are then extracted from this altered version of color image. Next, multiple support vector machines (SVMs) are trained with disjoint sets of gradient feature sets. Finally, probabilistic decisions on the test images are made using sigmoid estimation based posterior calculations on the ensemble bagging of SVMs. We demonstrate that this SVM-bagging system is capable of boosting the probability of the pixels near the contour regions compared to that of non-contour ones.
{"title":"Probabilistic contour mapping using oriented gradient features and SVM-bagging","authors":"Shubhra Aich, Yong-Cheol Lee, Chil-Woo Lee","doi":"10.1109/FCV.2015.7103732","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103732","url":null,"abstract":"In this paper, we propose a supervised approach to find out the probabilistic mapping of semantic contours in color images. We prepare a new color image modifying the RGB color planes to incorporate reasonable within-object contrasts in all the color planes. Color gradient based features are then extracted from this altered version of color image. Next, multiple support vector machines (SVMs) are trained with disjoint sets of gradient feature sets. Finally, probabilistic decisions on the test images are made using sigmoid estimation based posterior calculations on the ensemble bagging of SVMs. We demonstrate that this SVM-bagging system is capable of boosting the probability of the pixels near the contour regions compared to that of non-contour ones.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"34 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129156060","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 : 1900-01-01DOI: 10.1109/FCV.2015.7103731
Shiro Tanaka, Aki Takayanagi, M. Tsuchida, Y. Sakaguchi, Hiromi T. Tanaka
Material perception of woven fabrics is produced by the geometric factors based on microscopic structure of the fabric surface and optical factors based on reflection and transmission properties of fibers and yarns. In order to faithfully reproduce the material perception of the fabric, we observe the high-resolution multi-band high dynamic range (HDR) image of the reflection of fabric surface in one pixel (size: 5 ~ 7μm), and estimate the spectral reflectance for each pixel using the partial least squares (PLS) method. In PLS method, the training samples corresponding to the observed image and the spectral reflectance that was measured by the spectrophotometer are required. However, spectral reflectance of fabric is higher than training samples such as color patches. Therefore, the estimation accuracy decreased in the range of high luminance compared with the training samples. In this paper, we propose a method for estimating the spectral reflectance of having high reflectance, such as fabric, and show the effectiveness of the proposed method. In addition, we have confirmed that the specular reflection containing the object color is stronger on the high light area of the fabric.
{"title":"Estimating spectral reflectance of fabrics from high-resolution multi-band HDR images","authors":"Shiro Tanaka, Aki Takayanagi, M. Tsuchida, Y. Sakaguchi, Hiromi T. Tanaka","doi":"10.1109/FCV.2015.7103731","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103731","url":null,"abstract":"Material perception of woven fabrics is produced by the geometric factors based on microscopic structure of the fabric surface and optical factors based on reflection and transmission properties of fibers and yarns. In order to faithfully reproduce the material perception of the fabric, we observe the high-resolution multi-band high dynamic range (HDR) image of the reflection of fabric surface in one pixel (size: 5 ~ 7μm), and estimate the spectral reflectance for each pixel using the partial least squares (PLS) method. In PLS method, the training samples corresponding to the observed image and the spectral reflectance that was measured by the spectrophotometer are required. However, spectral reflectance of fabric is higher than training samples such as color patches. Therefore, the estimation accuracy decreased in the range of high luminance compared with the training samples. In this paper, we propose a method for estimating the spectral reflectance of having high reflectance, such as fabric, and show the effectiveness of the proposed method. In addition, we have confirmed that the specular reflection containing the object color is stronger on the high light area of the fabric.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"27 23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128677717","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 : 1900-01-01DOI: 10.1109/FCV.2015.7103721
Chi-Min Oh, Abdullah Nazib, Hao Xu, Joopil Moon, Chil-Woo Lee
In this paper we introduce a definition of centralization problem in multiple object tracking. The centralization problem occurs when more than two targets are contacting in close area. As a result, their trackers could be wrongly located in the center position of their targets. It is especially found when the appearances and shape of contacting targets are identical and the likelihood score for each tracker can be maximum value in their central position among their targets. After centralization problem their tracker could track wrong targets. Therefore we propose new interaction model to avoid centralization problem and our performance shows that almost all of centralization problem can be removed.
{"title":"Centralization problem of contacting interaction in multiple object tracking","authors":"Chi-Min Oh, Abdullah Nazib, Hao Xu, Joopil Moon, Chil-Woo Lee","doi":"10.1109/FCV.2015.7103721","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103721","url":null,"abstract":"In this paper we introduce a definition of centralization problem in multiple object tracking. The centralization problem occurs when more than two targets are contacting in close area. As a result, their trackers could be wrongly located in the center position of their targets. It is especially found when the appearances and shape of contacting targets are identical and the likelihood score for each tracker can be maximum value in their central position among their targets. After centralization problem their tracker could track wrong targets. Therefore we propose new interaction model to avoid centralization problem and our performance shows that almost all of centralization problem can be removed.","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":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130770376","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 : 1900-01-01DOI: 10.1109/FCV.2015.7103728
Shunsuke Shimizu, H. Koyasu, Yoshinori Kobayashi, Y. Kuno
3D object pose estimation is one of the most important challenges in the field of computer vision. A huge amount of image resources such as images on the web or photos taken before can be utilized if the system can estimate the 3D pose from a single image. On the other hand, the object's category and position on the image can be estimated by using state of the techniques in the general object recognition. We propose a method for 3D pose estimation from a single image on the basis of the known object category and position. We employ Regression Forests as the machine learning algorithm and HOG features as the input vectors. The regression function is created based on HOG features which express the differences in shapes depending on the viewing directions and corresponding poses. We evaluate the accuracy of pose estimation by using multiple objects with different categories.
{"title":"Object pose estimation using category information from a single image","authors":"Shunsuke Shimizu, H. Koyasu, Yoshinori Kobayashi, Y. Kuno","doi":"10.1109/FCV.2015.7103728","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103728","url":null,"abstract":"3D object pose estimation is one of the most important challenges in the field of computer vision. A huge amount of image resources such as images on the web or photos taken before can be utilized if the system can estimate the 3D pose from a single image. On the other hand, the object's category and position on the image can be estimated by using state of the techniques in the general object recognition. We propose a method for 3D pose estimation from a single image on the basis of the known object category and position. We employ Regression Forests as the machine learning algorithm and HOG features as the input vectors. The regression function is created based on HOG features which express the differences in shapes depending on the viewing directions and corresponding poses. We evaluate the accuracy of pose estimation by using multiple objects with different categories.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133180731","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 : 1900-01-01DOI: 10.1109/FCV.2015.7103703
Hao Liu, Atsushi Shimada, Xing Xu, H. Nagahara, Hideaki Uchiyama, R. Taniguchi
Making a reasonable ranking on images in dataset is one of the main objectives for object retrieval challenge, and in this paper we intend to improve the ranking quality. We follow the idea of query expansion in previous researches. Based on the use of bag-of-visual-words model, tf-idf scoring and spatial verification, previous method applied a pointwise style learning in query expansion stage, using but not fully exploring verification results. We intend to extend their learning approach for better discriminative power in retrieval. In re-ranking stage we propose a method using pairwise learning, instead of pointwise learning previously used. We could obtain more reliable ranking on a shortlist of examples. If this verification itself is reliable, a good re-ranking should best preserve this sub-ranking order. Thus in our proposed method, we are motivated to leverage a pairwise learning method to incorporate the ranking sequential information more efficiently. We evaluate and compare our proposed method with previous methods over Oxford 5k dataset, a standard benchmark dataset, where our method achieve better mean average precision and showed better discriminative power.
{"title":"Query expansion with pairwise learning in object retrieval challenge","authors":"Hao Liu, Atsushi Shimada, Xing Xu, H. Nagahara, Hideaki Uchiyama, R. Taniguchi","doi":"10.1109/FCV.2015.7103703","DOIUrl":"https://doi.org/10.1109/FCV.2015.7103703","url":null,"abstract":"Making a reasonable ranking on images in dataset is one of the main objectives for object retrieval challenge, and in this paper we intend to improve the ranking quality. We follow the idea of query expansion in previous researches. Based on the use of bag-of-visual-words model, tf-idf scoring and spatial verification, previous method applied a pointwise style learning in query expansion stage, using but not fully exploring verification results. We intend to extend their learning approach for better discriminative power in retrieval. In re-ranking stage we propose a method using pairwise learning, instead of pointwise learning previously used. We could obtain more reliable ranking on a shortlist of examples. If this verification itself is reliable, a good re-ranking should best preserve this sub-ranking order. Thus in our proposed method, we are motivated to leverage a pairwise learning method to incorporate the ranking sequential information more efficiently. We evaluate and compare our proposed method with previous methods over Oxford 5k dataset, a standard benchmark dataset, where our method achieve better mean average precision and showed better discriminative power.","PeriodicalId":424974,"journal":{"name":"2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV)","volume":"79 2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131593388","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：