On content-based image synthesis

We develop a content-based image synthesis software to compose a novel picture from many existing pictures. Text labels and user drawn sketches are provided to retrieve desired pictures for composing. We add text labels to images according to image features, so that our search by text labels can generate accurate results. Shape features are used to exclude pictures that do not match the user drawn sketches. We also improve the interactive image segmentation to be more automatic. We apply seamless blending techniques to compose pictures.