`V' shapes. The resulting object, seen from a suitable distance, has a rather curious speckled look to it. The artist Cannaletto made a whole heap of money out of his discovery that you could make green paint look a bit like water if you painted lots of white ripples on top of it. Similar results can be obtained by playing with the background stipple patterns on the control panel of a Macintosh computer. Such a result is called a texture and may be obtained, as in the case of the Mac, with binary images.
If we take a rectangular grid and move it around the image, we find that the pattern seen at one location strikingly resembles the pattern seen when it is shifted by some more or less fixed number of pixels in any fixed direction. In general, this pattern is statistical rather than deterministic, but we can extract some statistics for the values in windows in this case too. For example, on a piece of reflective steel, upon which some characters have been incised, there is no regularity apart from that in the characters, but the mean grey level and the variance may be fairly constant. The eye can quite easily distinguish between regions having the same mean grey level if one has a much higher variance than the other, providing the variation is within some kind of visual acuity. Similarly, other kinds of correlation between pixel values may appear as a `texture' variation
in the image, and this information may be extracted and put to use in obtaining a `feature' of an image. In the exercise at the end of chapter one where you were asked to distinguish pictures of underclad ladies from pictures of trees, even the most glorious of New England Fall Maple trees can easily be distinguished automatically from naked flesh, not by colour but by texture. The variation in a square block of 25 pixels is bigger for leaves than for flesh; try it if you don't believe me. I have studied such images with assiduous attention to detail and am confident of my ground.