Learn More
This paper presents an implementation of nonlinear anisotropic filtering, using resistive fuse circuitry, on the 64×64 cell MIPA4k mixed mode image processor array. Anisotropic filtering can be used to implement very effective edge preserving noise filtering, image segmentation as well as direct edge detection. The MIPA4k processor cell is capable of(More)
An evolved version of variable block-size analog motion estimation cell structure is presented. Due to the space restrictions of interconnects, used advanced variable block-size motion estimation architecture necessitates using deep-submicron design technology, which enables the dense spacing of interconnects. Therefore, a novel architecture realizing the(More)
In this paper we propose a self-adapting image sensing scheme that compresses a wide dynamic range scene so that important visual characteristics are preserved throughout the perceived dynamic range. The exposure times of sensors are made adaptive to local illumination conditions while signal is being integrated. Resistive filtering with locally adaptive(More)
Mathematical morphology provides powerful methods for image analysis and segmentation both for binary and grayscale images. By implementing the required basic functions directly in hardware, the advantage of using a locally connected massively parallel array processor structure, such as CNN, can be fully realized. This paper presents an improved method and(More)
In this paper we present a novel approach for implementing a local binary pattern (LBP) based feature extraction system with cellular nonlinear networks (CNNs). The LBP methodology is based on transforming local binary features of an image into micro-patterns that can be used to, for example, moving object detection and face recognition and detection. We(More)
In this paper we propose a dedicated hardware for extraction of local binary pattern (LBP) feature vectors. The LBP method transforms local features of image data into binary micro-patterns that represent local and global features of the image. The LBP method can be used in applications such as texture classification, moving object detection and face(More)
A novel method of shifting data within a cellular nonlinear network is presented. The method utilizes a 3-neighborhood connection and a novel cell type to reduce the error due to the shift. The novel cell type halves the amount of current mirrors needed for each shift. The applications of the method lie in block-based motion estimation of current video(More)
This paper introduces a 64x64 cell massively parallel processor array and measurement system. The approach taken in the chip is based on the inclusion of dedicated computing cores, each of which have been designed to implement a specific set of operations very efficiently. The chip can be roughly divided into two grayscale processing cores, one for order(More)
A combination selection based device mismatch calibration for mixed-mode array processors is discussed. Clear benefits in implementation area and accuracy, compared to large transistors can be reached by using mismatch calibration which is based on the combination selection of minimum-sized transistors. By utilizing the in-cell memory elements present in a(More)
A new simple method for shifting reference image data within an analog cellular array for variable block-size motion estimation is presented. The method leads to a greatly reduced number of neighborhood connections for each cell of the array, and allows for all shifts within the Marku, J., et al, (2007) search area to be performed in a single step, with(More)