Learning Spatial-Semantic Context with Fully Convolutional Recurrent Network for Online Handwritten Chinese Text Recognition
read more
Citations
Building fast and compact convolutional neural networks for offline handwritten Chinese character recognition
Accurate, data-efficient, unconstrained text recognition with convolutional neural networks
Aggregation Cross-Entropy for Sequence Recognition
Text Recognition in the Wild: A Survey
A signature-based machine learning model for distinguishing bipolar disorder and borderline personality disorder.
References
Deep Residual Learning for Image Recognition
Long short-term memory
Distinctive Image Features from Scale-Invariant Keypoints
Going deeper with convolutions
Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
Related Papers (5)
An End-to-End Trainable Neural Network for Image-Based Sequence Recognition and Its Application to Scene Text Recognition
Frequently Asked Questions (10)
Q2. What is the main advantage of the residual recurrent network?
The residual recurrent network of MC-FCRN not only substantially accelerates the convergence procedure but also promotes the performance significantly, while adding neither extra parameter nor computational burden to the system.
Q3. What is the reason why the recurrent network is not optimized?
This is because the recurrent network had not yet functioned during that period; thus FCN network can be optimized directly with CTC loss function through the residual connection.
Q4. What is the efficient method for OHCTR?
Among the above-mentioned methods, over-segmentation [1], [2], [3], [4], [5], i.e., an integrated segmentation-recognition method, is the most efficient method and still plays a crucial role in OHCTR.
Q5. What are the main arguments for the degradation problem in training very deep networks?
On the other hand, highway network [37] [38] and residual connection [22] [39] were advocated to solve the degradation problem [22] in training very deep networks.
Q6. What is the effect of the sliding window on the path signature?
Fig. 2d shows that, although connections are randomly added between adjacent strokes within a character or between characters, their impact on the path signature of the original input string is not significant, which proves that path signature based on sliding window has excellent local invariance and robustness.
Q7. What is the way to evaluate the effectiveness of the proposed system?
To evaluate the effectiveness of the proposed system, the authors conducted experiments on the standard benchmark dataset CASIA-OLHWDB [57] and the ICDAR2013 Chinese handwriting recognition competition dataset [58] for unconstrained online handwritten Chinese text recognition.
Q8. What is the k-th iterated integral of the signature?
As adjacent sampling points of text samples are connected by a straight line D = (D1t , D 2 t ) with t ∈ [a, b], the iterated integrals P (D)(k)a,b can be calculated iteratively as follows:P (D) (k) a,b ={ 1 , k = 0,(P (D) (k−1) a,b ⊗4a,b)/k , k ≥ 1,(5)where4a,b := Db −Da denotes the path displacement and ⊗ represents the tensor product.
Q9. What is the advantage of the residual recurrent network?
As a result, their residual recurrent network captures the contextual information from a sequence through the term∑L−1l=1 h(ql(x)) in an elegant manner, making the text recognition process more efficient and reliable than processing each character independently.
Q10. What is the proposed solution to the problem of a fully convolutional recurrent network?
The authors propose a new fully convolutional recurrent network (FCRN) for spatial context learning to overcome this problem by leveraging a fully convolutional network, a residual recurrent network, and connectionist temporal classification, all of which naturally take inputs of arbitrary size or length.