The Number of Effective Pixels for Digital Still Cameras and Its Picture Quality

To realize good total image quality for digital still cameras, it is important to achieve coexistence of higher resolution and lower noise level. In this research we develop the noise simulation model based on the noise properties of actual digital still cameras. Psychophysical experiments are conducted by using a series of simulated images which have different resolutions and noise levels. Contribution of resolution and noise level to total image quality are discussed by analyzing the experimental results. Introduction Although the image area size of CCD devices employed in consumer digital still cameras is getting smaller, the number of effective pixels is increasing dramatically for recent years. This is achieved by the advanced technology of fine electronic-device processing. However, at the same time, it is usually accompanied with some loss of light efficiency captured by the device and consequently increase of noise in the resultant images. Therefore, it is getting more important to investigate the relationships among the number of pixels, device noise level, and perceived image quality. Many researches have been already carried out to evaluate total image quality. R. Shaw proposed the criteria which satisfy simultaneous resolution and noise. G. M. Johnson has been studying to develop the image quality metrics by using a vision model. However, we think it is necessary to gather psychophysical data which examine the relationship among the attributes to the overall image quality. In this research, first we develop the noise model to estimate the noise properties propagated through the camera system. This model covers a wide range of variations including device-pixel size, ISO speed settings, etc. Then we make the image quality map so that we can learn the adequate effective pixel number of CCD devices to provide a preferable image under a given picture size and noise level. Noise Model Figure 1 shows a simple structure of digital still camera. Optical signal through a lens is sampled and transformed to electric singal by a CCD. After the signal is amplified by a analog processor, it is converted to digital signal by a A/D converter. By a digital processor this digital signal is processed to the digital image and it is recorded to a memory. Lens CCD Analog Processor Digital Processor A/D Memory Figure 1. Structure of digital still camera It is relatively easy to get the properties of noise propagation on digital process. For example, amplifying signal double increases S/N (signal to noise ratio) by 6 dB. And if signal is digital-filtered, S/N varies depending on the frequency response of the digital-filter. There are many kinds of independent noise which are generated throughout a digital still camera. However, we assume the total noise is composed of dark noise, fixed pattern noise and shot noise. Dark noise is mainly due to thermal noise. Fixed patten noise is due to the variation of aperture size etc., and it is generated during the manufacturing process of CCDs. Shot noise is due to the variation of photons and it is known that its magnitude is proportional to the square root of light intensity level. The total noise is calculated by taking the sum of those three individual noise and defined by the following equation 1. 2 2 2 s f d t N N N N + + = (1) where Nt, Nd, Nf and Ns denote total noise, dark noise, fixed pattern noise and shot noise respectively. We attempt to measure these three kinds of noise and get total noise. Dark Noise is measured on the condition that the incident light is intercepted. We measure it by varying analog gain, and get the result as shown in figure 2. It is found that the property of dark noise is proportional to analog gain. IS&T's 2003 PICS Conference