Industrial baking is a temperature and time controlled process, which considers neither the actual quality of the raw materials nor the process parameters like humidity, pastry temperature and actual pastry status. Furthermore the baking process is irreversible. Therefore, without a process monitoring considering the actual process state, suboptimal results may be achieved. To obtain optimal results, an automated monitoring system is required, but not yet available. Such a system must be able to identify the baking goods and the current state of the baking process represented by color and size of the baking goods. To develop such a system, digital image processing was used. An optical system was implemented, which was able to make digital images of the baking goods from inside the oven in a continuous form. The goal was the development of algorithms for distinction of baking goods and characterization of color saturation and shape, altogether resulting in an optical online process monitoring system. By using a modified Viola-Jones algorithm the kind of baking good in the oven is identified with an error of 5.6%. The error of automated determination of the width and height change of bread rolls with respect to manual evaluation is less than 4%. Based on a neural network, the baking good is identified pixel by pixel. The training error of the neuronal net was 7.0%. This allows the calculation of the evolution of lightness and color saturation. Using this information, the state of the baking process is identified reliably. Therefore, the basics for the automatic control is provided. (C) 2012 Elsevier Ltd. All rights reserved.