The use of standard color reference targets at image acquisition allows to compensate for different camera characteristics, illumination conditions and exposure times, ensuring true colors in digital photo workflows. Reliable automatic detection of reference targets makes color correction faster, and this becomes critical in mass digitization processes. The existing automatic algorithms usually assume that there is little perspective distortion and/or that the scanning resolution is known, achieving very limited results for example when the relative size of the color target is unknown. In this paper we present a preprocessing step that aims at automatically detecting a region of interest (ROI) where the reference target is located. We compare the performance of one of the available automatic tools (CCFind) with and without this preprocessing step, and show a considerable improvement in the detection of color reference targets in a new challenging dataset. In addition, a simple template matching approach is compared with the performance of CCFind. The results show that the selection of a smaller ROI complements well with the existing approaches and helps to improve detection.