Abstract
The aim of this thesis is to better understand recognition of old handwritten weather logs. This process should be able to recognize the handwriting in these documents with a high-level of accuracy. While Optical Character Recognition (OCR) has existed for years, there are still short comings. These methods frequently confuse letters and numbers when it comes to handwritten texts. Since OCRs struggle with recognizing handwritten characters, more robust recognition methods were employed by use of Convolutional Neural Networks (CNN). Further exploration is done by using a compound network, the joining of a CNN with a Long Short-Term Memory (LSTM) network. Each of these models are put to the test with detailed experimentation and comparative analysis. While all models perform with high accuracy, the compound network performed faster and with high accuracy than the lone CNN.