Matching Software-Generated Sketches to Face Photographs With a Very Deep CNN, Morphed Faces, and Transfer Learning

Galea, Christian; Farrugia, Reuben A.

doi:10.1109/tifs.2017.2788002

Cited by 61 publications

(56 citation statements)

References 45 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We find that the increase in the number of convolution layers brings no positive results to accuracy. This may be due to the degradation of accuracy when there are too many convolution level, and happened in variety of DL studies . Based on this conclusion, we conduct our experiments only with single convolution layer in following steps.…”

Section: Experiments and Resultsmentioning

confidence: 95%

See 1 more Smart Citation

Sensor‐based activity recognition independent of device placement and orientation

Shi

Zuo

Zhang

et al. 2020

Trans Emerging Tel Tech

View full text Add to dashboard Cite

Human activity recognition (HAR) is a prominent subfield of pervasive computing and also provides context of many applications such as healthcare, education, and entertainment. Most wearable HAR studies assume that sensing device placement and orientation are fixed and never change. However, this condition is actually not always guaranteed in the real scenario and recognition result is influenced by the distortion as consequence. To handle this, our work proposes a new model based on convolutional neural network to extract robust features which are invariant of device placement and orientation, to train machine learning classifiers. We first carry out experiments to show negative effects of this problem. Then, we apply the convolutional neural network-based hybrid structure on the HAR. Results show that our method provides 15% to 40% accuracy promotion on public data set and 10% to 20% promotion on our own data set, both with distortion.Trans Emerging Tel Tech. 2020;31:e3823.wileyonlinelibrary.com/journal/ett

show abstract

Section: Experiments and Resultsmentioning

confidence: 95%

“…Deep learning algorithm has been paid great attention in recent years. It has excellent performance in many fields such as image processing, natural language processing, and data mining . Several practical and efficient DL network structures have been designed in former research like AlexNet and Resnet .…”

Section: Related Workmentioning

confidence: 99%

Sensor‐based activity recognition independent of device placement and orientation

Shi

Zuo

Zhang

et al. 2020

Trans Emerging Tel Tech

View full text Add to dashboard Cite

show abstract

“…These datasets include 123 photos for the AR dataset and the corresponding composite sketch using FACES and Indntikit software, respectively. The Uom-SGFS dataset [9] contains 1200 images from the Color FERET datasets and the corresponding viewed software-generated composite sketches. There are two parts in the Uom-SGFS dataset, one is created using the EFIT-V software, and the other one adopts image editing software to make the sketch image closer to the corresponding face photo.…”

Section: Implementation and Experimental Resultsmentioning

confidence: 99%

“…We compared the performance with the ones in [10], [14], [9] for each dataset. For Uom-SGFS datasets, since the attribute for each face sketch is very different from the corresponding face photo, the recognition accuracy is lower than the state of art one.…”

Section: Figure 2 Recognition Accuracy Of Proposed Methods Using Thrementioning

confidence: 99%

An Improved Siamese Network for Face Sketch Recognition

Liu

Hou

2019

2019 International Conference on Machine Learning and Cybernetics (ICMLC)

View full text Add to dashboard Cite

Face sketch recognition identifies the face photo from a large face sketch dataset. Some traditional methods are typically used to reduce the modality gap between face photos and sketches and gain excellent recognition rate based on a pseudo image which is synthesized using the corresponded face photo. However, these methods cannot obtain better high recognition rate for all face sketch datasets, because the use of extracted features cannot lead to the elimination of the effect of different modalities' images. The feature representation of the deep convolutional neural networks as a feasible approach for identification involves wider applications than other methods. It is adapted to extract the features which eliminate the difference between face photos and sketches. The recognition rate is high for neural networks constructed by learning optimal local features, even if the input image shows geometric distortions. However, the case of overfitting leads to the unsatisfactory performance of deep learning methods on face sketch recognition tasks. Also, the sketch images are too simple to be used for extracting effective features. This paper aims to increase the matching rate using the Siamese convolution network architecture. The framework is used to extract useful features from each image pair to reduce the modality gap. Moreover, data augmentation is used to avoid overfitting. We explore the performance of three loss functions and compare the similarity between each image pair. The experimental results show that our framework is adequate for a composite sketch dataset. In addition, it reduces the influence of overfitting by using data augmentation and modifying the network structure.

show abstract

“…In recent years, a convolution neural network (CNN) has had excellent performance in image and video recognition, [22][23][24] recommender system, 25,26 and nature language processing. 27,28 Compared to the traditional machine learning methods, which learn the hand-crafted features, [29][30][31] CNN is independent from prior knowledge and human effort in feature design.…”

Section: Methodsmentioning

confidence: 99%

Cerebral micro‐bleeding identification based on a nine‐layer convolutional neural network with stochastic pooling

Wang

Sun

Mehmood

et al. 2019

Concurrency and Computation

View full text Add to dashboard Cite

Cerebral micro-bleedings are small chronic brain hemorrhages caused by structural abnormalities of the small vessels. CMBs can be found from individuals with stroke at memory clinics and even healthy elderly people. CMBs indicate hemorrhage-prone pathological states. Research shows that CMBs are associated with an increased risk of future ischemic stroke, intra-cerebral hemorrhage (ICH), dementia, and death. Considering that CMBs severely influence people's life, it is necessary to identify the CMBs in an early stage to prevent from further deterioration and to help people live a healthy life. In this paper, we proposed using CNN with stochastic pooling for the CMB detection. CNN has good performance in image and video recognition, recommender system, and nature language processing. Based on the collected subject, the experiment result shows that the six-convolution layer and three fully-connected layer CNN, nine-layers in total, achieved sensitivity, specificity, accuracy, and precision as 97.22%, and 97.35%, 97.28%, and 97.35% in average of ten runs, which shows better performance than five state-of-the-art methods. KEYWORDScerebral micro-bleeding, convolution neural network, detection, intra-cerebral hemorrhage, stochastic pooling, stroke INTRODUCTIONCerebral micro-bleedings (CMBs) are small chronic brain hemorrhages, which can be caused by structural abnormalities of the small vessels. CMBs can be found from individuals with stroke and memory clinics that is initially set up for research purposes. 1 CMBs indicate hemorrhage-prone pathological states. Research shows that CMBs are associated with an increased risk of future ischemic stroke, intra-cerebral hemorrhage (ICH), dementia, and death. 2 Previous studies show that, in a range of 18% to 68%, patients with ischemic stroke/transient ischemic attack have a high prevalence of CMBs. There are 23% of prevalence in patients with first-ever ischemic stroke and 44% in patients with recurrent ischemic stroke reported by a previous systematic review. 3 The strong relationship with CMBs and increased risk of ICH suggests that the antithrombotic drugs Concurrency Computat Pract Exper. 2020;32:e5130. wileyonlinelibrary.com/journal/cpe to test the performance of the proposed structure of CNN based on the stochastic pooling, we compared different pooling methods, including average pooling, max pooling, and stochastic pooling. Meanwhile, we compared CNN with different convolution number and the state-of-the-art method for CMB detection. The experiment result shows that our designed structure of CNN based on the stochastic pooling achieved the best performance in sensitivity as 97.22%, specificity as 97.35%, accuracy as 97.28%, and precision as 97.35%.The limitation of our work is that we may need to move the window from top to bottom and from left to right. This is time-consuming. It is better to use concurrent convolution in the future studies, and we may need to replace the fully connected layers with conv layers. Another limitation is that we have two labels (CMB and ...

show abstract

Matching Software-Generated Sketches to Face Photographs With a Very Deep CNN, Morphed Faces, and Transfer Learning

Cited by 61 publications

References 45 publications

Sensor‐based activity recognition independent of device placement and orientation

Sensor‐based activity recognition independent of device placement and orientation

An Improved Siamese Network for Face Sketch Recognition

Cerebral micro‐bleeding identification based on a nine‐layer convolutional neural network with stochastic pooling

Contact Info

Product

Resources

About