This artificial retina chip, consisting of an array of Variable Sensitivity Photo-detector cells (VSPD), attains versatile focal plane image processing using vector-matrix multiplication. We designed an n-MOS VSPD with a pn-photodiode and a differential amplifier, which realizes programmable posltive and negative sensitivity. The pixels of the fabricasted 32×32 pixel artificial retina chip are 60×60μm^2 with 2.5% aperture ratio in a 2μm n-MOS process. The photosensitivity is 0.56μ A/lx for 1ms accumulation time. We demonstrate image capture in "video mode" and "edge extraction mode"
We investigate a new computational image sensor which compresses image signal in the process of image acquisition. We fabricated a prototype compression sensor in a 2μm CMOS technology and examine the prototype chip. We also design a new compression sensor, in which power consumption is improved and circuits for controlling rate and address is implemented.
Massively parallel proceasing vision with photo-detectors and processing elements directly connected can realize high speed operation of obtaining image and processing it much faster than conventional systems using CCD camera. In this paper, we describe a general purposed parallel processing vision system SPE (Sensory Processing Element)-4k, algorithms for such high speed vision system and high speed visual feedback system as its application in order to show the effectiveness of massively parallel processing vision.
We proposed a new man-machine Interface using a hand gesture recognition system. The system consists of a CCD camera, an image processing unit, a Quantizer Neuron Chip (QNC) which we have already developed, a personal computer and a display. Users of the system can control virtual buttons on the display by means of hand gesture which is recognized by QNC. We applied the system to some tasks such as object transportation in virtual rooms or controling a TV and a video.
Optical Flow estimation is useful to describe the velocity information of motion image relative to the observer, and gradient based method is often used. In this method, we must choose a suitable weighting factor determined by the average brightness level and noise in the image, denoted by α^2. In this paper, we proposed to vary α^2 according to the local brightness level and noise in the image, and parallel analog image processors for motion detection circuitry based variable method we designed α^2. Using this method, we can improve the accuracy of motion estimation of any condition.