A System on FPGA for Fast Handwritten Digit Recognition in Embedded Smart Cameras

This paper presents the first FPGA SoC implementation solution to the hand-written multi-digit numbers recognition problem. Our proposed solution employs a novel digit extraction method which relies on the identification of images' non-zeros columns instead of the widely used computationally-expensive segmentation method. Digit prediction is performed by a multi-layer neural network. The paper presents a design and an FPGA implementation of the proposed solution; and also discusses various optimization techniques in the neural network implementation that lead to increased performance. Our proposed solution achieves a 96.76% detection accuracy and up to 2.47x speed-up in comparison to software solutions.

[1]  Yoshua Bengio,et al.  Gradient-based learning applied to document recognition , 1998, Proc. IEEE.

[2]  Tomaso A. Poggio,et al.  Regularization Theory and Neural Networks Architectures , 1995, Neural Computation.

[3]  Christophe Bobda,et al.  Sparsely connected neural networks in FPGA for handwritten digit recognition , 2016, 2016 17th International Symposium on Quality Electronic Design (ISQED).

[4]  Salim Ouchtati,et al.  Segmentation and Recognition of Handwritten Numeric Chains , 2007 .

[5]  Gaetano Borriello,et al.  Paper-Digital Workflows in Global Development Organizations , 2015, CSCW.

[6]  Michelle Effros,et al.  Improved bounds for the rate loss of multiresolution source codes , 2003, IEEE Trans. Inf. Theory.

[7]  B Saldanha Luca,et al.  Sparsely connected neural networks in FPGA for handwritten digit recognition , 2016 .

[8]  Guang-Bin Huang,et al.  Extreme Learning Machine for Multilayer Perceptron , 2016, IEEE Transactions on Neural Networks and Learning Systems.

[9]  Yaroslav Bulatov,et al.  Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks , 2013, ICLR.

[10]  Tom Schaul,et al.  No more pesky learning rates , 2012, ICML.

[11]  Christophe Bobda,et al.  A System on Reconfigurable Chip for Handwritten Digit Recognition , 2015, 2015 IEEE 23rd Annual International Symposium on Field-Programmable Custom Computing Machines.

[12]  Thomas G. Dietterich.  Overfitting and undercomputing in machine learning , 1995, CSUR.