Skip to Main Content
In this paper we examine the scalability of several implementations of the 2-dimensional discrete cosine transformation in the context of image processing. By scaling down the quality of the transformation the required computational complexity also decreases. Using several benchmark images we can show that no significant loss of image quality results from downscaling the computational complexity by up to 60%. This property can be used to switch between different quality levels during the execution of the DCT. A low quality level is used if only few time remains to finish the computation; otherwise a higher quality level can be used. For a certain execution model we show that this switching between quality levels can be used to meet the real-time demands of the executed image processing application even in the presence of a permanent fault in the execution units.