Handwritten digits recognition using multiple instance learning

Now more and more heterogeneous handwritten digits data sets appear into sight. But traditional handwritten digits recognition algorithms are usually based on the homomorphism data sets. For solving the problem that handwritten digits data sets of different feature spaces can´t compute, we constructed heterogeneous handwritten digits representation model based on multiple instance learning (MIL) where a bag contains handwritten digits data from different feature spaces. Handwritten digits classification algorithms (HB and HeterMIL) are designed and compared for handwritten digits recognition. Experiment results confirmed that the heterogeneous handwritten digits data representation model and recognition algorithms can solve the heterogeneous handwritten digits recognition effectively.