| This thesis addresses text detection in images utilizing the properties of the Discrete Cosine Transform (DCT) coefficients as a discrimination statistic. By modeling the distribution of the DCT coefficients, a statistical method is proposed. A Maximum Likelihood (ML) estimator is used to estimate the model parameter and a chi2 test is employed in order to measure how well the DCT coefficients match the modeling. Based on the spectral energy property, the sum of the absolute value of the DCT coefficients in each block is considered as an energy statistic for detection. Linear Discriminant Analysis (LDA) is conducted to calculate the optimal threshold. In order to strengthen the discrimination statistic, weights are employed, which include uniform, binary, linear and quadratic weights. In addition, a filtering approach is introduced into the system. Finally, supervised learning methods operating on the DCT coefficient are utilized. By introducing the LDA into the modified Support Vector Machine (SVM), better classification is achieved. An evaluation of the statistical and energy based discrimination algorithms is conducted in the Receiver Operating Characteristics (ROC) space. The ROC curves show that there is a tradeoff between the True Positive Rate (TPR) and the False Positive Rate (FPR) for all weight configurations. In terms of maximizing the separation between two distributions, the experimental results show that the quadratic weighted energy achieves the best recall and precision. Considering the misclassification rate, the modified SVM achieves the best performance. |
