2.3.5. Other Visual Phenomena

    It is well known that a sharper image generally looks even more pleasant to a human viewer than an original image. This is often exploited in improving the appearance of an image for a human viewer. It is also a common experience that an unnatural aspect catches a viewer’s attention. A po.sitive aspect of this phenomenon is that it can be exploited in such applications as production of television commercials. A negative aspect is that it sometimes makes it more difficult to develop a successful algorithm using computer processing techniques. For example, some image processing algorithms are capable of reducing a large amount of background noise. In the process. however, they introduce noise that has an artificial tone to it. Even when the amount of the artificial noise introduced is much less than the amount by which the background noise is reduced, the artificial noise may catch a viewer’s attention more, and a viewer may prefer the unprocessed image over the processed image.

The visual phenomena discussed in the previous sections can be explained simply. at least at a qualitative level; however, many other visual phenomena cannot be explained simply, in part due to our lack of knowledge. For example. a visual phenomenon that involves a fair amount of central level processing cannot be explained in a satisfactory manner. Figure 2.24 shows a sketch consisting of just a small number .of lines. How we can associate this image with Einstein is not clear. The example does demonstrate, however, that simple outlines representing the gross features of an object are important for its identification. This can be exploited in such applications as object identification in computer vision and the development of a very low bit-rate video telephone system for the deaf.

The visual phenomena discussed above relate to the perception of light that shines continuously. When light shines intermittently, our perception depends a great deal on its frequency. Consider a light that flashes on for a brief duration N times per second. When N is small, the light flashes are perceived to be separate.

As we increase N, an unsteady flicker that is quite unpleasant to the human viewer occurs. As we increase N further, the flicker becomes less noticeable, and eventually the observer can no longer detect that the light intensity is changing as a function of time. The frequency at which the observer begins perceiving light flashes as continuous light is called the critical flicker frequency or fusion frequency. The fusion frequency increases as the size and overrallintensitv of the flickering object increase. The fusion frequency can be as low as a few cycles/sec for a very dim, small light and may exceed 100 cycles/sec for a very bright, larger light. When a flicker is perceived, visual acuity is at its worst.

Intermittent light is common in everyday vision. Fluorescent lights do not shine continuously, as they appear to, but flicker at a sufficiently high rate (over 100 times/see) that fusion is reached in typical viewing conditions. Avoiding the perception of flicker is an important consideration in deciding the rate at which a CRT (cathode ray tube) display monitor is refreshed. As is discussed further in Section 2.4, CRT display monitors are illuminated only for a short period of time. For an image to be displayed continuously without the perception of flicker, the monitor has to be refreshed at a sufficiently high rate. Typically, a CRT display monitor is refreshed 60 times per second. With 2:1 interlace, which is discussed further in Section 2.4. this corresponds to 30 frames/sec. The current NTSC (National Television Systems Committee) television system employs 30 frames/sec with 2:1 interlace. For motion pictures. 24 frames per second are shown, with one frame shown twice. The effective flicker rate is therefore 48 frames/sec. In addition, the typical viewing condition in a cinema is very dark. decreasing the fusion frequency to below 40 cycles/sec. For this reason, flickering is not visible in a motion picture, even though the screen is dark for approximately half of the time.

    Even though each frame of a motion picture or television broadcast is actually still, and only a finite number of frames are shown in a second, the objects in the scene appear to be moving in a continuous manner. This effect, known as motion rendition, is closely related to the phi phenomenon. Consider two pulsating light sources separated by approximately I degree of an observer’s viewing angle. When the lights shine for one msec each with a separation of 10 msec. the light is perceived to move continuously from one source to the other. When the time difference between the two lights is on the order of I msec, they appear to flash simultaneously. When the time difference is more than I second. they are perceived as two separate flashes. This is known as the phi phenomenon.

    In general, frame rates that are sufficiently high to avoid flicker are adequate for motion rendition. The fact that an object appears to move in a continuous manner in a motion picture or television broadcast does not necessarily imply that the sampling rate along the temporal dimension is above the Nyquist rate. For objects with sufficiently rapid motion, sampling the temporal dimension 24 times. sec or 30 times/sec is far lower than the Nyquist rate. and temporal aliasng occurs. Temporal aliasing does not always cause motion discontinuity. In a movie, we sometimes see a wheel that moves continuously, but backwards. In this case, motion rendition is present. but significant temporal aliasing has occurred. Our current knowledge of the flicker effect, motion rendition, the temporal aliasing effect. and their interrelations is far from complete. A comprehensive understanding of this topic would be useful in a number of applications, such as bit rate reduction by frame elimination in a sequence of image frames.