Clothing attribute recognition has become an essential technology,which enables users to automatically identify the characteristics of clothes and search for clothing images with similar attributes.However,existing me...Clothing attribute recognition has become an essential technology,which enables users to automatically identify the characteristics of clothes and search for clothing images with similar attributes.However,existing methods cannot recognize newly added attributes and may fail to capture region-level visual features.To address the aforementioned issues,a region-aware fashion contrastive language-image pre-training(RaF-CLIP)model was proposed.This model aligned cropped and segmented images with category and multiple fine-grained attribute texts,achieving the matching of fashion region and corresponding texts through contrastive learning.Clothing retrieval found suitable clothing based on the user-specified clothing categories and attributes,and to further improve the accuracy of retrieval,an attribute-guided composed network(AGCN)as an additional component on RaF-CLIP was introduced,specifically designed for composed image retrieval.This task aimed to modify the reference image based on textual expressions to retrieve the expected target.By adopting a transformer-based bidirectional attention and gating mechanism,it realized the fusion and selection of image features and attribute text features.Experimental results show that the proposed model achieves a mean precision of 0.6633 for attribute recognition tasks and a recall@10(recall@k is defined as the percentage of correct samples appearing in the top k retrieval results)of 39.18 for composed image retrieval task,satisfying user needs for freely searching for clothing through images and texts.展开更多
In order to obtain a better sandstone three-dimensional (3D) reconstruction result which is more similar to the original sample, an algorithm based on stationarity for a two-dimensional (2D) training image is prop...In order to obtain a better sandstone three-dimensional (3D) reconstruction result which is more similar to the original sample, an algorithm based on stationarity for a two-dimensional (2D) training image is proposed. The second-order statistics based on texture features are analyzed to evaluate the scale stationarity of the training image. The multiple-point statistics of the training image are applied to obtain the multiple-point statistics stationarity estimation by the multi-point density function. The results show that the reconstructed 3D structures are closer to reality when the training image has better scale stationarity and multiple-point statistics stationarity by the indications of local percolation probability and two-point probability. Moreover, training images with higher multiple-point statistics stationarity and lower scale stationarity are likely to obtain closer results to the real 3D structure, and vice versa. Thus, stationarity analysis of the training image has far-reaching significance in choosing a better 2D thin section image for the 3D reconstruction of porous media. Especially, high-order statistics perform better than low-order statistics.展开更多
A new sub-pixel mapping method based on BP neural network is proposed in order to determine the spatial distribution of class components in each mixed pixel.The network was used to train a model that describes the rel...A new sub-pixel mapping method based on BP neural network is proposed in order to determine the spatial distribution of class components in each mixed pixel.The network was used to train a model that describes the relationship between spatial distribution of target components in mixed pixel and its neighboring information.Then the sub-pixel scaled target could be predicted by the trained model.In order to improve the performance of BP network,BP learning algorithm with momentum was employed.The experiments were conducted both on synthetic images and on hyperspectral imagery(HSI).The results prove that this method is capable of estimating land covers fairly accurately and has a great superiority over some other sub-pixel mapping methods in terms of computational complexity.展开更多
The paper proposes a new method of "Separated Same Rectangle Feature (SSRF)" for face detection. Generally, Haar-like feature is used to make an Adaboost training algorithm with strong classifier. Haar-like featur...The paper proposes a new method of "Separated Same Rectangle Feature (SSRF)" for face detection. Generally, Haar-like feature is used to make an Adaboost training algorithm with strong classifier. Haar-like feature is composed of two or more attached same rectangles. Inefficiency of the Haar-like feature often results from two or more attached same rectangles. But the proposed SSRF are composed of two separated same rectangles. So, it is very flexible and detailed. Therefore it creates more accurate strong classifier than Haar-like feature. SSRF uses integral image to reduce execuive time. Haar-like feature calculates the Sanl of intmsities of pixels on two or more rectangles. But SSRF always calculates the stun of intensities of pixels on only two rectangles. The weak classifier of Ariaboost algorithm based on SSRF is fastex than one based on Haar-like feature. In the experiment, we use 1 000 face images and 1 000nm- face images for Adaboost training. The proposed SSRF shows about 0.9% higher acctwacy than Haar-like features.展开更多
Aimed at the problems of infrared image recognition under varying illumination,face disguise,etc.,we bring out an infrared human face recognition algorithm based on 2DPCA.The proposed algorithm can work out the covari...Aimed at the problems of infrared image recognition under varying illumination,face disguise,etc.,we bring out an infrared human face recognition algorithm based on 2DPCA.The proposed algorithm can work out the covariance matrix of the training sample easily and directly;at the same time,it costs less time to work out the eigenvector.Relevant experiments are carried out,and the result indicates that compared with the traditional recognition algorithm,the proposed recognition method is swift and has a good adaptability to the changes of human face posture.展开更多
In view of the important prestige of Zhang Heng’s seismoscope in world culture,scholars in different countries have performed restoration research,however,so far all restored models have only reached the drawing desi...In view of the important prestige of Zhang Heng’s seismoscope in world culture,scholars in different countries have performed restoration research,however,so far all restored models have only reached the drawing design and exhibition stage. In recent decades,new research achievements on historical materials,archaeology and historical seismology have been made in Chinese academic society,which lay a good foundation for the establishment of a new restored model. Not only does the new model more accord with historical materials,but it can be used to detect earthquakes for the first time. This paper will introduce the restoration in four aspects: scientific train of thought,structural restoration,formation restoration and scientific testing.展开更多
An algorithm to refine and clean gait silhouette noises generated by imperfect motion detection techniques is developed,and a relatively complete and high quality silhouette is obtained.The silhouettes are sequentiall...An algorithm to refine and clean gait silhouette noises generated by imperfect motion detection techniques is developed,and a relatively complete and high quality silhouette is obtained.The silhouettes are sequentially refined in two levels according to two different probabilistic models.The first level is within-sequence refinement.Each silhouette in a particular sequence is refined by an individual model trained by the gait images from current sequence.The second level is between-sequence refinement.All the silhouettes that need further refinement are modified by a population model trained by the gait images chosen from a certain amount of pedestrians.The intention is to preserve the within-class similarity and to decrease the interaction between one class and others.Comparative experimental results indicate that the proposed algorithm is simple and quite effective,and it helps the existing recognition methods achieve a higher recognition performance.展开更多
Distance effect has been regarded as the best established marker of basic numerical magnitude processes and is related to individual mathematical abilities. A larger behavioral distance effect is suggested to be conco...Distance effect has been regarded as the best established marker of basic numerical magnitude processes and is related to individual mathematical abilities. A larger behavioral distance effect is suggested to be concomitant with lower mathematical achievement in children. However, the relationship between distance effect and superior mathematical abilities is unclear. One could get superior mathematical abilities by acquiring the skill of abacus-based mental calculation (AMC), which can be used to solve calculation problems with exceptional speed and high accuracy. In the current study, we explore the relationship between distance effect and superior mathematical abilities by examining whether and how the AMC training modifies numerical magnitude processing. Thus, mathematical competencies were tested in 18 abacus-trained children (who accepted the AMC training) and 18 non-trained children. Electroencephalography (EEG) waveforms were recorded when these children executed numerical comparison tasks in both Arabic digit and dot array forms. We found that: (a) the abacus-trained group had superior mathematical abilities than their peers; (b) distance effects were found both in behavioral results and on EEG wave- forms; (c) the distance effect size of the average amplitude on the late negative-going component was different between groups in the digit task, with a larger effect size for abacus-trained children; (d) both the behavioral and EEG distance effects were modulated by the notation. These results revealed that the neural substrates of magnitude processing were modified by AMC training, and suggested that the mechanism of the representation of numerical magnitude for children with superior mathematical abilities was different from their peers. In addition, the results provide evidence for a view of non-abstract numerical representation.展开更多
基金National Natural Science Foundation of China(No.61971121)。
文摘Clothing attribute recognition has become an essential technology,which enables users to automatically identify the characteristics of clothes and search for clothing images with similar attributes.However,existing methods cannot recognize newly added attributes and may fail to capture region-level visual features.To address the aforementioned issues,a region-aware fashion contrastive language-image pre-training(RaF-CLIP)model was proposed.This model aligned cropped and segmented images with category and multiple fine-grained attribute texts,achieving the matching of fashion region and corresponding texts through contrastive learning.Clothing retrieval found suitable clothing based on the user-specified clothing categories and attributes,and to further improve the accuracy of retrieval,an attribute-guided composed network(AGCN)as an additional component on RaF-CLIP was introduced,specifically designed for composed image retrieval.This task aimed to modify the reference image based on textual expressions to retrieve the expected target.By adopting a transformer-based bidirectional attention and gating mechanism,it realized the fusion and selection of image features and attribute text features.Experimental results show that the proposed model achieves a mean precision of 0.6633 for attribute recognition tasks and a recall@10(recall@k is defined as the percentage of correct samples appearing in the top k retrieval results)of 39.18 for composed image retrieval task,satisfying user needs for freely searching for clothing through images and texts.
基金The National Natural Science Foundation of China(No.60972130)
文摘In order to obtain a better sandstone three-dimensional (3D) reconstruction result which is more similar to the original sample, an algorithm based on stationarity for a two-dimensional (2D) training image is proposed. The second-order statistics based on texture features are analyzed to evaluate the scale stationarity of the training image. The multiple-point statistics of the training image are applied to obtain the multiple-point statistics stationarity estimation by the multi-point density function. The results show that the reconstructed 3D structures are closer to reality when the training image has better scale stationarity and multiple-point statistics stationarity by the indications of local percolation probability and two-point probability. Moreover, training images with higher multiple-point statistics stationarity and lower scale stationarity are likely to obtain closer results to the real 3D structure, and vice versa. Thus, stationarity analysis of the training image has far-reaching significance in choosing a better 2D thin section image for the 3D reconstruction of porous media. Especially, high-order statistics perform better than low-order statistics.
基金Sponsored by the National Natural Science Foundation of China(Grant No. 60272073, 60402025 and 60802059)by Foundation for the Doctoral Program of Higher Education of China (Grant No. 200802171003)
文摘A new sub-pixel mapping method based on BP neural network is proposed in order to determine the spatial distribution of class components in each mixed pixel.The network was used to train a model that describes the relationship between spatial distribution of target components in mixed pixel and its neighboring information.Then the sub-pixel scaled target could be predicted by the trained model.In order to improve the performance of BP network,BP learning algorithm with momentum was employed.The experiments were conducted both on synthetic images and on hyperspectral imagery(HSI).The results prove that this method is capable of estimating land covers fairly accurately and has a great superiority over some other sub-pixel mapping methods in terms of computational complexity.
基金supported by the Korea Research Foundation Grant funded by the Korean Government(MOEHRD),the MKE(The Ministry of Knowledge Economy,Korea)the ITRC(Information Technology Research Center)support program(NIPA-2009-(C1090-0902-0007))
文摘The paper proposes a new method of "Separated Same Rectangle Feature (SSRF)" for face detection. Generally, Haar-like feature is used to make an Adaboost training algorithm with strong classifier. Haar-like feature is composed of two or more attached same rectangles. Inefficiency of the Haar-like feature often results from two or more attached same rectangles. But the proposed SSRF are composed of two separated same rectangles. So, it is very flexible and detailed. Therefore it creates more accurate strong classifier than Haar-like feature. SSRF uses integral image to reduce execuive time. Haar-like feature calculates the Sanl of intmsities of pixels on two or more rectangles. But SSRF always calculates the stun of intensities of pixels on only two rectangles. The weak classifier of Ariaboost algorithm based on SSRF is fastex than one based on Haar-like feature. In the experiment, we use 1 000 face images and 1 000nm- face images for Adaboost training. The proposed SSRF shows about 0.9% higher acctwacy than Haar-like features.
基金Sponsored by the Natural Science Fund of Heilongjiang province(Grant No. F2007-13)Science and Technology Research Projects in Office of Education of Heilongjiang province(Grant No.11531034)the Heilongjiang Postdoctoral Science Foundation(Grant No.LBH-Z06054)
文摘Aimed at the problems of infrared image recognition under varying illumination,face disguise,etc.,we bring out an infrared human face recognition algorithm based on 2DPCA.The proposed algorithm can work out the covariance matrix of the training sample easily and directly;at the same time,it costs less time to work out the eigenvector.Relevant experiments are carried out,and the result indicates that compared with the traditional recognition algorithm,the proposed recognition method is swift and has a good adaptability to the changes of human face posture.
基金sponsored by the National Natural Science Foundation of China(40644019)the Special Found of Scientific Research Program for "The Optimization and Design for Reconstruction Models of Seismoscope",China Earthquake Administration
文摘In view of the important prestige of Zhang Heng’s seismoscope in world culture,scholars in different countries have performed restoration research,however,so far all restored models have only reached the drawing design and exhibition stage. In recent decades,new research achievements on historical materials,archaeology and historical seismology have been made in Chinese academic society,which lay a good foundation for the establishment of a new restored model. Not only does the new model more accord with historical materials,but it can be used to detect earthquakes for the first time. This paper will introduce the restoration in four aspects: scientific train of thought,structural restoration,formation restoration and scientific testing.
基金the National Natural Science Foundation of China (No. 60675024)
文摘An algorithm to refine and clean gait silhouette noises generated by imperfect motion detection techniques is developed,and a relatively complete and high quality silhouette is obtained.The silhouettes are sequentially refined in two levels according to two different probabilistic models.The first level is within-sequence refinement.Each silhouette in a particular sequence is refined by an individual model trained by the gait images from current sequence.The second level is between-sequence refinement.All the silhouettes that need further refinement are modified by a population model trained by the gait images chosen from a certain amount of pedestrians.The intention is to preserve the within-class similarity and to decrease the interaction between one class and others.Comparative experimental results indicate that the proposed algorithm is simple and quite effective,and it helps the existing recognition methods achieve a higher recognition performance.
基金supported by the National High-Tech R&D Program(863)of China(Nos.2012AA011603 and 2012AA011602)the National Natural Science Foundation of China(Nos.30900389 and 31270026)
文摘Distance effect has been regarded as the best established marker of basic numerical magnitude processes and is related to individual mathematical abilities. A larger behavioral distance effect is suggested to be concomitant with lower mathematical achievement in children. However, the relationship between distance effect and superior mathematical abilities is unclear. One could get superior mathematical abilities by acquiring the skill of abacus-based mental calculation (AMC), which can be used to solve calculation problems with exceptional speed and high accuracy. In the current study, we explore the relationship between distance effect and superior mathematical abilities by examining whether and how the AMC training modifies numerical magnitude processing. Thus, mathematical competencies were tested in 18 abacus-trained children (who accepted the AMC training) and 18 non-trained children. Electroencephalography (EEG) waveforms were recorded when these children executed numerical comparison tasks in both Arabic digit and dot array forms. We found that: (a) the abacus-trained group had superior mathematical abilities than their peers; (b) distance effects were found both in behavioral results and on EEG wave- forms; (c) the distance effect size of the average amplitude on the late negative-going component was different between groups in the digit task, with a larger effect size for abacus-trained children; (d) both the behavioral and EEG distance effects were modulated by the notation. These results revealed that the neural substrates of magnitude processing were modified by AMC training, and suggested that the mechanism of the representation of numerical magnitude for children with superior mathematical abilities was different from their peers. In addition, the results provide evidence for a view of non-abstract numerical representation.