New member II
Joined: 21 Aug 2012
|Posted: 02/11/2015, 03:39 pm Post subject: Some More about Imagery
|First a word of clarification . The word “spatial” is often used, in some publications, to refer to imagery taken from space, typically from an earth observation satellite. This is not what will be discussed here.
In an earlier essay, I outlined the radiometric domain of an image, usually, in a digital form, quantified as the gray levels ( plus color levels in a color image) of the individual pixels ( from picture elements) .
We’ll now move on to look at some basic radiometric enhancement concepts and begin also considering information contained within the pixel matrix of rows and columns ( typically “x” and “y” axes ) whilst examining some of the terminology and processes involved in their analysis and interpretation.
Firstly, let us recap on the diagrammatic illustration of a digital display at the end of the “Radiometric Domain” essay.
Fig.1 Schematic digital array.
The array can be seen to be arranged as a matrix of pixels (often referred to as a raster display) each with its own digital number (DN), more commonly known as its radiometric, radiance, “brightness” or gray scale value, which is the figure in each square (i.e. pixel) represented in the diagram . The pixel positions within the matrix are identified by their coordinates in the “x” (column) and “y”( row) values, the imagery convention being ( x, y).
The coordinates number from the top left of the image, the first being (1,1) the next to the right (2,1) and so on. N.B. Confusingly some graphics conventions start at (0,0,)
The third (z) axis in the digital imagery construct is the brightness value which is usually expressed in quantification of bits e.g. 8-bit (256 values) ,however, in this axis the values would be normally expressed as 0-255.
Physical Elements of the Digital Image.
With most hand-held cameras an array of small sensors is situated in the “image plane”, which is where the film would be exposed in an analog camera. However, some observation satellite and other families of imaging sensors do use somewhat different systems which are separate cases for interpretation, not covered here.
This array of sensors, called a Charge Coupled Device or CCD, contains a great number of tiny semiconductors, each equating to a pixel in the output image. The size of the array is often expressed in Megapixels ( MP or millions of pixels ). An 8MP digital camera can, theoretically at least, capture twice as much information as a 4MP camera with an otherwise similar imaging system. When this figure is multiplied by the number of gray levels, digital images often result in very large files, the number of bytes being usually represented in increments of 210 (1,024) or more:
1 Kilobyte (KB) = 1,024 bytes
1 Megabyte (MB) = 1,024 KB
1 Gigabyte (GB) = 1,024 MB
1 Terabyte (TB) = 1,024 GB
With the increasing capability of computers to handle such ever growing file sizes, it is now possible to perform quite sophisticated imagery calculations on a PC, which even a large mainframe computer would have struggled with a few years ago.
As an aside, it should be noted that a CCD which is physically bigger can have bigger semiconductors, which in turn can potentially gather up many more photons and give a better quality pixel result. An analogy of this in analog silver film imaging is the good quality of some very old images , obtained when both lens and film technology was less sophisticated. The reason for the quality was simply the large size of the photographic plates, and later the film formats. Strategic aerial reconnaissance cameras used the same concept for obtaining high quality imagery of distant targets, often by using film formats of 9 or even 18 inches – or more, with appropriate lens systems.
Each semiconductor in creating its pixel gathers up photons from a minute square section of the scene viewed through the lens and converts it into a tiny electric charge proportional to the incoming energy hitting it during the brief period of exposure. All of these charges are then, extremely rapidly, stripped off of the CCD row by row, in a manner often likened to a line of men operating a fire bucket brigade, and dumped into a file in the storage medium ( typically a chip) , usually along with additional metadata ( i.e. data describing data) that are supplied by the camera.
Typically both the fineness of the spatial data (N.B. MP) and the radiometric resolution of the pixels far exceed the limitations of the human visual system. In other words there is far more information contained within the digital image than can be directly perceived by the observer. For example, very approximately, the human eye has angular spatial resolution characteristics which enables detection of objects subtending around one minute ( i.e. 1/60 deg) of arc. In simple physiological terms one degree of a scene is projected across 288µm of the retina by the eye's lens and the number of tiny sensors (rods and cones) contained within this value defines a physical constraint along with measured responses. Radiometrically, only around 50 brightness levels can be differentiated. There are variations due to different contrast/illumination and duration parameters but the basic rules apply fairly well in empirical considerations of both radiometric and spatial domains.
It should be noted that the number of output display pixels from a CCD device is usually much greater than the number of pixels ( screen pixels) on a computer screen , each screen pixel typically being an average value of several display pixels covering its position in the screen matrix. Incidentally, this process is typically reversed by the computer when a part of the original output image is enlarged on the screen. Enlarging beyond the point where the original output pixels just become discernable on the screen, however, is generally a wasted exercise in direct visual interpretation.
To enable aspects of this information contained within a digital image to be made visually perceptible to the observer there are various processes that are applied to make this information more apparent. We will start by considering how the data in the radiometric domain may be made more apparent or “enhanced”.
Density Slicing and Contrast Enhancement
In Fig.1 above we saw how an individual pixel is identified by its unique “x” ,”y” and “z” coordinates in an image. If we now imagine these pixels hypothetically as just being like a collection of marbles in a bag with their gray levels written on them, we could collect them all up and in the case of a 0-255 gray level spread count them into discrete groups, identified by their gray level values . These would come out with a distribution of gray level values from 0 to 255. We can represent these in a number of ways, one of which is commonly used is a distribution histogram .
Fig 2.Example of a Distribution Histogram.
In this diagram the pixel gray levels can be seen to be distributed in a pattern with most pixels in the middle of the gray levels ( “mid gray”) dropping away to fewer pixels to the left ( towards zero or “black” ) and similarly to the right towards saturated or “white). The red line is a curve drawn through the histogram levels and is referred to as a distribution curve.
So if the overall curve is displaced to the left you have a dark image and if to the right a light image. In addition, if a large number of pixels have a similar gray level the distribution will be a bit like the Empire State Building in profile. With the pixels bunched so close together in brightness level the human visual system, with its limitation of only being able to discern around 50 Brightness levels, will not be able to discern many levels of gray which are contained within the image.
This practical example will perhaps illustrate the point:
Fig 3. Digital Image of a Submarine and Histogram of Conning Tower Area. .
It can be seen that there is a bunching of the pixel values around about the 80 – 100 gray level range ( the dotted curve is just a cumulative representation of the same data). It can also be seen that it is not easy to pick out detail on the hull around the conning tower as the eye cannot readily detect the differing gray levels in such an electronic “soft copy” screen display, or “tonal” values in a “hard copy” image version such as a print.
However if we now take a slice of the distribution histogram in the area of the peak of gray values (density slicing) and pull it out across the “x” axis , ( the yellow colored histogram values below ) we keep the relative values of gray level intact but open them up across the distribution histogram, increasing the separation levels between them to a point at which the human eye can now differentiate . The image has been contrast enhanced.
Fig 4. Enhanced Submarine Image and Matching Histogram.
N.B. No additional information has been added to the image. The information already held within the image has merely been re-displayed to aid the observer in visualisation. This is the essence of Imagery Enhancement , including all of the routines developed and implemented in IPACO.
As well as the radiometric ( z) domain, imagery enhancement routines are also employed to display less clear data in the spatial (x, y,) domain ( i.e across the image in any and all directions). To understand the basis of these routines it is helpful to first of all understand the concept of spatial frequency.
When considering the term “frequency”, the first idea that comes to mind is of some repeating variable with time, such as for a example 50hz electicity mains frequency, or a “high frequency train service”.
With imagery we use the concept of spatial frequency to refer to how the brightness values in pixels change as we scan across the image. So in simple terms we are looking at changes in radiometry ( or gray level) compared to units of measurement across the image.
In this image of pebbles on a beach, note how the rapid changes in gray level across the scene give it a rough “texture”. A scan across a row or column of the image pixels would give a result somewhat similar to the diagram on the right.
Fig.5 High Spatial Frequency Dominant Image.
In comparison, the image of clouds below shows predominantly gradual changes in gray level across the scene, giving it a “ smooth” texture. With a scan along a row or column of pixels in this case giving a result somewhat similar to the diagram on the right.
Fig.6 Low Spatial Frequency Dominant Image.
In practice, images comprise a large and complex mixture of high and low spatial frequency components and the data within these components can be extracted and analysed by tools working in the spatial frequency domain.
The basic theory and use of these tools in imagery enhancement will be outlined in part (ii) of this article.