A respective model of background color is determined for each image in a set of images. For each of the images, a respective background sub-image is extracted from the image, the background sub-image is segmented into regions, and each of the regions is labeled with a respective color label. An image query is generated from a target color template. For each of the images, a respective score is calculated from the image query and the respective color model. At least one of the images is retrieved from an image collection based on the respective scores.