For machine vision applications, image sensors with high-spatial resolution are not always required or desirable though the low-resolution image is not sufficient to analyze the details. However, since the amount of data is reduced, the computation can be performed fast which is critical in real-time applications. In addition, the unnecessary detail information can be filtered out in low-resolution image

In order to utilize all advantages of various resolution images, we have proposed the analog-to-digital conversion structure that is a type of column-parallel multi-bit sigma-delta oversampling ADC using the controlled reset scheme. We expect that the proposed CMOS active pixel sensor with spatial-temporal multi-resolution readout can be applied to prevailing hierarchical image processing algorithms.

Figure1 shows the effect of multi-resolution readout. The first is captured image with normal 8-bit 30 fps operation. There are motion blur in moving area. The second is 4-bit 150 fps snapshot image. Although motion blur is significantly reduced, large spatial noise is noticeable. The final image is multi-resolution image. Pixels in moving area is merged spatially, and background image is same with 8-bit normal image.

Figure 1. Sample Images