Baseline Detection for On-line Cursive Thai Handwriting Recognition Based on PCA

In the handwriting recognition, the baseline detection is one of the preprocessing processes. It is important for the efficiency of the handwriting recognition. For Thai language, a Thai word composes of vowels and tones located on above or below the consonants. This is different from other languages. Therefore, detecting the baseline of Thai handwritten words becomes a challenging task. In this paper, we propose a new approach to detect the baseline of Thai handwritten words using PCA algorithm. The principle of this method is to remove the outliers from words before conducting PCA algorithm to find the baseline direction. For finding the outliers, the proposed method finds the centroid of the handwritten word using three different methods including the mean method, the median method and the middle method. We evaluate the performance of the proposed technique using a real world dataset collected from various Thai natives. From the experiments, the mean method is the best among the three methods. Then, we compare the performance of baseline detection between the mean method and the ordinary PCA approach. The result shows that the proposed approach outperforms the ordinary PCA approach. The number of words which the proposed approach is able to detect baseline more accurately is 21 from 30 words.