This dataset is composed of 6.000 CD/DVD cover images and some associated labels. The term “cover” refers to the font-facing panel of a CD/DVD package, and, increasingly, the primary image accompanying a digital download of the album, or of its individual tracks.

These images have been downloaded from using a Java custom application making use of the Amazon API (

CD covers represent an interesting challenge related to several computer vision and pattern recognition problems. In the present state of the dataset, labels are related to regions of the image where we can find printed text and it is designed for studying the problem of unconstrained text detection in complex backgrounds.