Bootstrapping for efficient handwritten digit recognition

Multidimensional multistage k-NN classifiers for handwritten digit recognition

“…If we look for the k nearest patterns of a level only among some classes, the computational cost is reduced quite more. In [14] a bootstrapping algorithm to improve the classification accuracy and some algorithms to reduce the training set size are applied. The results obtained are compared with Hart's condensing algorithm and we can observe that when they reduce the training set size above Hart's condensing training set size, the hit rate is reduced a lot.…”

Pattern Synthesis Using Multiple Kernel Learning for Efficient SVM Classification

“…The ordinary bootstrapping is a method of resampling the given data and has been a successful method for error estimation [15][16][17][18]. The bootstrapping method that creates (not selects) new training samples was proposed by H a m a m o t o et al [1] that acts as a smoother of the distribution of the training samples and was successfully applied in the design of 1NN classifier, particularly in high dimensional spaces.…”

“…The bootstrapping method that creates (not selects) new training samples was proposed by H a m a m o t o et al [1] that acts as a smoother of the distribution of the training samples and was successfully applied in the design of 1NN classifier, particularly in high dimensional spaces. Further, H a m a m o t o et al [1] generated bootstrap samples by combining the training data locally and illustrated that the NNC (Nearest Neighbour Classifier) based on bootstrap patterns performed better than that of K-NNC (K-nearest-neighbor classifier) based on the original data [18].…”

“…The bootstrapping technique has the ability to remove outliers which therefore reduces the variability in the data, as well as removes noise. This in turn increases the distance between two close patterns belonging to different classes and thereby improves the generalization performance of the classifier [18].…”

Handwritten Digit Classification Based on Alpha-Beta Associative Model

“…Handwritten digits recognition is an important research area. Several techniques have been used to improve the recognition performance: contours [1], bootstrapping [2], neural networks [3], Singular Value Decomposition (SVD) [4], k-Nearest Neighbor (k-NN), Learning Vector Quantization (LVQ), Support Vector Classifiers (SVC) with Radial Base Function (RBF) kernel [5], Multi-Layer Perceptron (MLP) and LeNet [6], and holographic associative memories [7], among others.…”