Realize handwritten numeral recognition (data set 50000 pictures) Compare 3 kinds of algorithm neural network, gray average value, SVM respective accuracy rate-jason NIU

50000 pictures of the handwritten data set 0~9 recognition of Arabic numerals, and the accuracy of the analysis of the results,

Handwritten digital data set download: http://yann.lecun.com/exdb/mnist/

First of all, using the properties of the picture itself, the image of the gray average to identify the classification, I run out of the accuracy rate is about 22%

Using the gray average of images to classify and realize handwritten image recognition (DataSet 50000 photos)--jason Niu

Secondly, using the SVM algorithm, I run out of the accuracy rate is about 93%, the specific code please click

SVM: Using SVM algorithm to realize handwritten image recognition (data set 50000 pictures)-jason Niu

Finally, using the deep learning neural network, I run out of the accuracy rate is about 94%, the specific code please click

NN: Using neural networks of deep learning to realize handwritten numeral recognition (data set 50000 photos)-jason Niu

Finally, we find that the algorithm learning quality of neural network and SVM is very high, while the traditional gray-scale average algorithm is not satisfactory!

Realize handwritten numeral recognition (data set 50000 pictures) Compare 3 kinds of algorithm neural network, gray average value, SVM respective accuracy rate-jason NIU