Handwritten Digit Recognition Application Based on Improved Naive Bayes Method

Handwritten digit recognition has a wide range of application scenarios, but because handwritten digits have the characteristics of randomness and great variability, it is often difficult to obtain high classification accuracy. This paper improves on the basis of the Naive Bayes classification algorithm, mainly including two aspects of work. On the one hand, in the data preprocessing stage, the local adaptive threshold method is used to perform the binarization of the sample data, so that the feature values of the sample data participating in the training are more accurate. On the other hand, considering the large number of features after image binarization, in order to prevent conditional probability multiplication from generating underflow, and also to improve the efficiency of algorithm execution, each feature conditional probability is multiplied by the sample size, and the floating-point operation was changed to integer operation in probability solving process. In addition, the operation result was increased by 1 as smooth to avoid the problem of zero probability. The method proposed in this paper is compared with the related improved algorithm that introduces logarithmic operation. The execution efficiency has been significantly improved, and the classification accuracy has also been improved.