The Media Processing Engine (MPE) is suitable for image recognition and other multimedia processing. It is a very long instruction word (VLIW) machine that can issue multiple instructions simultaneously. It consists of Toshiba's proprietary low-power 32-bit RISC core called a Media Embedded Processor (MeP) and a coprocessor suitable for multimedia processing called Image Recognition VLIW Coprocessor 2 with FPU (IVC2F). IVC2F can execute one or two SIMD (single instruction stream, multiple data stream) instructions, which perform one or two operations on multiple data in parallel. The IVC2F can efficiently perform double-precision floating-point operations. The Visconti Family can execute multiple applications simultaneously* because of its multi-core configuration containing multiple MPEs.
In addition to the Toshiba-original luminance-based CoHOG, the image recognition accelerator contained in the Visconti4 processes four kinds of feature quantities: color-based CoHOG, Co-occurrence Histograms of Pairs of Edge Orientations and Color Differences (CoHED), Co-occurrence Histograms of Color Differences (CoHD), and Color Histograms. This accelerator provides high recognition performance even at nighttime and at scenes with less luminance differences between objects and the background. The Visconti4 realizes nighttime pedestrian recognition as reliable as daytime pedestrian recognition available with conventional vision systems.