An effective mesh-generation strategy for image representation using data-dependent triangulation

A new mesh-generation method for image representation based on data-dependent triangulations (DDTs) is proposed. The proposed method is shown to produce meshes of higher quality (both in terms of mean squared error and subjectively) than those generated by two competing approaches, namely the greedy-point removal scheme of Demaret and Iske and a DDT-based scheme of Garland and Heckbert. Furthermore, our method is shown to achieve these excellent results at a relatively modest computational and memory cost.