Entropy information‐based heterogeneous deep selective fused features using deep convolutional neural network for sketch recognition

Hayat, Shaukat; She, Kun; Shahzad, Sara; Suwansrikham, Parinya; Mateen, Muhammad; Yu, Yang

doi:10.1049/cvi2.12019

Cited by 7 publications

(4 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In [33] presented a scheme for sketch recognition. It generates a discriminative features representation as a result of integrating asymmetry essential information from deep features.…”

Section: Related Workmentioning

confidence: 99%

“…Transfer Learning [26] 72.50% -Entropy-DCNNs [33] 72.93% -Double-channel CNN [32] 73.24% -Pre-Trained Model [34] 74% -Deformable-CNN [28] 79.10% -CNG-SCN [29] 80.10% -Dynamic Landmarks [27] 82.95% -Hybrid-CNN [30] 85.07% -Dense-CNN [31] 85.55% -DSSA [23] 79.47 % 68.00 % ResNet18 [41] 83.97 % 83.30 % ResNet50 [41] 86.03 % 90.75 % TCNet [17] 86.79 % 91.30 % Sketch-BERT [40] 88.30 % 91.40 % Our proposed FNN 98.07 % 96.69 % in dealing with the increase in data volume effectively.…”

Section: 87%mentioning

confidence: 99%

See 1 more Smart Citation

A Proposed Method to Acquire More Geometric Features from Hand-Drawn Sketches

2022

IJIES

View full text Add to dashboard Cite

We introduce a novel method for learning geometric features of hand-drawn sketches. The method is based on two models: the first depends on a newly-designed algorithm based on the Decision Tree (DT) technique; while the other is based on the Feed-forward Neural Network (FNN) technique. Each model consists of training and testing phase, we extract features of object and build the knowledge base in training phase. In testing phase, we test the learning ability to appear advantages of using DT with new objects and FNN with large data. The results of our method show desirable performance in learning. Experiments on TU-Berlin, QuickDraw, and our own dataset reveal the effectiveness of the method. We achieve learning accuracy 99.98% on our own dataset, 98.07% on TU-Berlin, and 96.69% on QuickDraw. Experiments signify that geometric feature representation and manipulation by our method brings about a substantial improvement over state-of-the-art methods on sketch classification.

show abstract

“…In [33] presented a scheme for sketch recognition. It generates a discriminative features representation as a result of integrating asymmetry essential information from deep features.…”

Section: Related Workmentioning

confidence: 99%

Section: 87%mentioning

confidence: 99%

A Proposed Method to Acquire More Geometric Features from Hand-Drawn Sketches

2022

IJIES

View full text Add to dashboard Cite

show abstract

“…The algorithm's facial recognition probability can be reduced from 98.6% to 4.8% by simulation experiments. Literature [17] in order to solve the complex problem of sketch recognition, a new sketch recognition method is proposed for five pre-trained deep convolutional networks fine-tuned for feature extraction, optimized fusion of multi-layer features with entropybased domain analysis, sorting features and outputting support vector machine for classification. The capability of the algorithm is evaluated on two sketch datasets, TU-Berlin and Sketchy, for scheme classification and retrieval tasks, and the functionality of the algorithm is optimized to a certain extent compared to the traditional.…”

Section: Introductionmentioning

confidence: 99%

Research and realization of computer image recognition system based on digital projection technology

2024

Applied Mathematics and Nonlinear Sciences

View full text Add to dashboard Cite

In this paper, the image Gabor features extracted by Gabor wavelet are fused with the image grayscale map to construct the enhanced Gabor features, and then combined with the characteristics of Gabor wavelet and convolutional layer, the Gabor feature extraction module, parallel convolution module and spatial transformation pooling module are designed. The corresponding Gabor convolutional layer and Gabor convolutional neural network are constructed using the appropriate module in accordance with the image recognition task application scenario. The convex set projection image super-resolution reconstruction method is used in this paper to improve the resolution of images with low resolution. The construction of a computerized image recognition system involves combining a Gabor convolutional neural network and a convex set projection method. This system has been tested and found to have a recognition accuracy of 93.5% for object images. This system’s ability to accurately recognize low-resolution shadow-obscured face images is possible thanks to using the convex set projection method to reconstruct the image and recognize it accurately with an accuracy of up to 93.85%. This system’s recognition performance for complex images has been proven through experiments.

show abstract

“…Early sketch-recognition methods mainly draw on frameworks from traditional object recognition approaches to extract handcrafted features such as HOG [5], GF-HOG [6], SIFT [7] etc., but their recognition accuracy is poor. With the develop-ment of deep convolutional neural networks (DCNN), many DCNN-based methods have emerged in the field of sketch recognition [8][9][10][11][12][13][14][15][16][17][18][19][20][21]. These methods are mainly divided into two categories: one is to directly transfer models pre-trained on ImageNet such as AlexNet [10] and LeNet [11] for sketchrecognition tasks; the other is to design a dedicated model framework for sketches [9,13,14,17,18,20].…”

Section: Introductionmentioning

confidence: 99%

Four‐branch Siamese network based on sketch‐specific data augmentation for sketch recognition

Wang

Zhao

Tan

2022

IET Image Processing

View full text Add to dashboard Cite

The shortened abstract is as follows: Sketch recognition has become an important hotspot issue because of sketch's intuitiveness and visualization. The existing sketch-recognition methods based on handcrafted features and deep features are insufficient in the recognition of the local information of sketches, and the recognition accuracy is not ideal. Accordingly, this paper proposes a four-branch Siamese network based on sketch-specific data augmentation to generate discriminative feature representations and improve the sketch-recognition accuracy. A sketch is an ordered list of strokes, we adopt the semantic information of strokes as the decomposition criteria to divide a sketch into three disjoint local blocks, and then combine the local blocks in pairs to form three new sketches. In order to give full play to the positive effect of local blocks on category prediction and enhance the fine-grained capability of the network, three newly generated sketches and the original sketch are combined to construct a four-branch Siamese network. Each branch network adopts the Sketch-A-Net architecture with the fully connected layer removed as the basic network, and we improve it by adding shortcut connection layer and multi-scale weighted bilinear coding (MWBC) modules. Compared with the state-of-the-art methods, the experimental results on the TU-Berlin dataset demonstrate the excellent performance of our model.

show abstract

Entropy information‐based heterogeneous deep selective fused features using deep convolutional neural network for sketch recognition

Cited by 7 publications

References 37 publications

A Proposed Method to Acquire More Geometric Features from Hand-Drawn Sketches

A Proposed Method to Acquire More Geometric Features from Hand-Drawn Sketches

Research and realization of computer image recognition system based on digital projection technology

Four‐branch Siamese network based on sketch‐specific data augmentation for sketch recognition

Contact Info

Product

Resources

About