Sketch and Match: Scene Montage Using a Huge Image Collection

We present a sketch-based approach to find matching source images for seamless image composition, by leveraging a large amount of image corpus collected from the Internet. Given a target image where the user draws a rough sketch to indicate the desired object fill-in, our system automatically searches a large image database, and returns a sparse set of matching images. These matching images contain salient regions semantically similar to the usersupplied sketch. Once the user has selected the preferred source region, it will be seamlessly pasted onto the target image where the sketch is drawn.