A CMOS Image Sensor for Focal Plane Decomposition
Zhiqiang Lin, Marci Hoffman, Walter D. Leon-Salas, Nathan Schemm, S. Balkar · 2005
An alternative image decomposition method that exploits prediction via nearby pixels has been integrated on a CMOS sensor focal plane. The proposed focal plane decomposition is compared to the 2D discrete wavelet transform (DWT) decomposition commonly used in state of the art compression schemes such as EBCOT and SPIHT. A technique for focal plane prediction is described that can be used to decompose the image into pyramid subbands as is done in DWT based approaches. Although its simulated compression performance is somewhat lower than that obtained using wavelet decomposition, prediction decomposition has much lower computational complexity and is suitable for implementation directly on the focal plane in a completely pixel parallel structure. When coupled with SPIHT-like encoding, the focal plane prediction technique obtains all the advantages of the DWT based SPIHT compression algorithm, including both high image quality at low bit rates and optimized data ordering for progressive image transmission. A CMOS test chip to perform single layer image decomposition based on pixel level parallel computation has been prototyped. The design and test results for this chip are discussed.