Image Retrieval for Augmented Reality


This project aimed at retrieving and localizing objects of interest from pictures taken by a smart phone. We have clean poster images in the dataset. Given an query image taken by a camera or a smartphone, we want to find the corresponding images in the dataset based on the content of the image. The technique can be used for product search, content based information retrieval/indexing.

Figure 1. Movie poster retrieval

Figure 2 shows examples of poster images taken by a smartphone

Figure 2. Example of poster images taken by a smartphone

Problems we want to solve

Problem 1. Feature representation local patterns are not efficient in discriminating images which have similar local appearance.

Figure 3. Local context helps to discriminate local features

Problem 2. Same local features have different importance in different images. Figure 4 shows an example. The descriptor in the wild is less important than the grass descriptor in a book cover.

Figure 4. Importance of the same local feature is different in different images