Extracting and indexing relevant information with computer vision algorithms in very large solar image archives allows investigating solar activity from a new perspective. Using computer vision algorithms, we have developed methods that work with very compact and concise descriptions of images. We apply our method to images from the Solar Dynamics Observatory (SDO) and present a proof-of-concept Query by Example (QBE) system. In addition we introduce a benchmark dataset, on one hand to evaluate our system, and on the other hand to allow comparisons of our results with other QBE systems in this domain.