Promising new directions have been opened up for content-based visual retrieval in recent years. Object-based retrieval which allows users to manipulate video objects as part of their searching and browsing interaction, is one of these. It is the purpose of this thesis to constitute itself as a part of a larger stream of research that investigates visual objects as a possible approach to advancing the use of semantics in content-based visual retrieval. The notion of using objects in video retrieval has been seen as desirable for some years, but only very recently has technology started to allow even very basic object-location functions on video. The main hurdles to greater use of objects in video retrieval are the overhead of object segmentation on large amounts of video and the issue of whether objects can actually be used efficiently for multimedia retrieval. Despite this, there are already some examples of work which supports retrieval based on video objects. This thesis investigates an object-based approach to content-based visual retrieval. The main research contributions of this work are a study of shot boundary detection on compressed domain video where a fast detection approach is proposed and evaluated, and a study on the use of objects in interactive image retrieval. An object-based retrieval framework is developed in order to investigate object-based retrieval on a corpus of natural image and video. This framework contains the entire processing chain required to analyse, index and interactively retrieve images and video via object-to-object matching. The experimental results indicate that object-based searching consistently outperforms image-based search using low-level features. This result goes some way towards validating the approach of allowing users to select objects as a basis for searching video archives when the information need dictates it as appropriate.