The quality of multicrystalline Silicon (mc-Si) solar cells strongly depends on the quality of the wafer material. Although carrier lifetime and resistivity measurements provide basic material properties, photoluminescence (PL) images on as-cut wafers provide a much deeper insight as many crystallization defects become visible. In this work, a feature-based classification framework is introduced to rate the quality of mc-Si wafers in the as-cut stage according to the expected IV parameters of the final solar cells. For the classification, three levels of complexity are compared. In addition to frequently used PL-image features, such as dislocations and contaminated regions, physically relevant image structures are described on a wavelet basis. Beyond the improved correlation of the PL-features with quality information, the more detailed description of image structures forms a basis to understand deviations of measured and expected material quality. The classification model is evaluated within a large experiment on more than 1000 wafers, including a broad variety of wafers from different ingots and bricks from five different manufacturers, which have been processed in an industrial production line to standard solar cells with aluminium back-surface field. It is demonstrated that the presented approach allows the open circuit voltage to be predicted with a mean absolute error (MAE) of only 1.1 mV if the training of the model is performed on a random set of wafers. Moreover, the quality of wafers from an unknown ingot can be predicted with an MAE of 1.7 mV and from an unknown manufacturer still with an MAE of 3.6 mV, which proves the actual strength of the chosen approach.