Content Based Image Retrieval

Content-based Image Retrieval (CBIR) is to use one or more image features to build an image indexing structure. Then given a query image, we can quickly retrieve a list of similar product images.

1. Image features: We can adopt four kinds of image feature: color-based (e.g., Histogram), texture-based (e.g., Local Binary Pattern), shape-based, and mixed (e.g., Joint Composite Descriptor). The more efficient and independent features we adopt, the higer retrieving accuracy we obtain. However, if we adopt too many features, especially high-dimensional features, we should spend more time in getting the retrieval result.

2. Image indexing: Efficient image indexing algorithms (e.g., Locality-Sensitive Hashing) can dramatically decrease the retrieval time.