Two phenomena clearly indicate that perceived brightness is not a simple function of intensity. The first phenomenon is based on the fact that visual systems tend to have "undershoot" and "overshoot" phenomena at the boundary of different intensity regions. The second phenomenon, called simultaneous contrast, relates to the fact that the brightness of the sensing area is not simply determined by its intensity.
The color of a person's perceived object is determined by the nature of the light reflected by the object. A light without color is called a monochromatic light or colored light. The only attribute of monochromatic light is its strength or size. Since the intensity of the perceptual monochromatic light varies from black to gray, and finally to white, gray level is often used to denote the intensity of monochromatic light. Monochrome images are often called grayscale images.
In addition to the frequency, there are three basic quantities to describe the quality of the colored light source: luminous intensity, luminous flux, and brightness. Luminous intensity-the total amount of energy flowing out of the light source; luminous flux-the energy that the observer perceives from the light source; brightness-the subjective depiction of light perception is not actually measurable.
Although imaging is based primarily on the energy emitted by electromagnetic waves, this is not the only way to generate images. Acoustic waves reflected by objects can also be used to form ultrasound images. Other major digital image sources are electron microscopy electron beams and synthetic images for graphics and visualization.
1. Capturing images using a single sensor
The single sensor that we are most familiar with should be a light diode whose output voltage waveform is proportional to the incident light. An optical filter is used to improve selectivity in front of the sensor.
In order to produce a two-dimensional image using a single sensor, there must be relative displacements in the X and y directions between the sensor and the imaging area. The film of a single sensor is mounted on a drum, and the mechanical drive of the drum provides a dimensional displacement. A single sensor mounted on the guide screw provides a displacement perpendicular to the drive direction, an inexpensive method for obtaining high-resolution images (but slower). Another similar mechanism uses a flat bed, and the sensor moves linearly in two directions.
2. Get the image using the stripe sensor
Compared to a single sensor, the more commonly used geometry is a sensor band consisting of embedded sensors that provide an imaging unit in one Direction. The motion perpendicular to the sensor band is imaged in the other direction.
A ring-shaped sensor is used for medical and industrial imaging to obtain a profile image of a three-dimensional object. The rotating X-ray source provides irradiation, and the sensor opposite the source of the radiation collects X-allegedly energy through the object. Note that the output of the sensor must be processed by the reconstruction algorithm, and the purpose of the reconstruction algorithm is to transform the perceptual data into a meaningful profile image.
3. Capturing images using a sensor array
A large number of electromagnetic waves and some ultrasonic sensing devices are often arranged in array form. This is also the main arrangement we see in the digital cameras. Its main advantage is to get a complete picture by focusing energy on the surface of the array. It is clear that the motion of the sensor arrangement discussed earlier is not required.
4. Simple image Forming Model
The image is represented by a two-dimensional function such as f (x, y). The physical meaning of F is determined by the image source. F must be non-zero and finite, and f can be characterized by two components: (1) The total amount of light that is exposed to the observed scene, and (2) the total amount of light reflected by the object in the scene. These two components are called the incident and reflection components, respectively, as I (x, y) and R (x, y), where I (x, y) from 0 to positive infinity, R (x, y) between 0 (fully absorbed) to 1 (full reflection). The nature of I (x, y) depends on the irradiation source, and the nature of R (x, y) depends on the characteristics of the imaging object. F = i*r This representation can also be used to illuminate the light through the media to form an image, but in this case, we will use the transmission coefficient instead of the reflection function.
When f represents the intensity or grayscale of a monochrome image at any coordinate, all intermediate values of F are shades of gray that vary from black to white.
2.4 Image Sampling and quantification
The output of most sensors is a continuous voltage waveform, and the amplitude and spatial characteristics of these waveforms are related to perceived physical phenomena. In order to produce a digital image, we need to convert continuous perceptual data into digital form. This conversion consists of two treatments: sampling and quantification.
2.4.1 Sampling and Quantification
The x and Y coordinates and amplitude of an image may be continuous. Digitizing the coordinate values is called sampling and digitizing the amplitude is called quantization. In addition to the discrete coefficients used, the quantization achieved by Kyoto relies heavily on the noise of sampled signals. It is clear that the quality of digital images depends largely on the number of samples and the grayscale level used in sampling and quantification. However, when selecting these parameters, the image content is an important consideration.
2.4.2 Digital Image representation
There are three basic methods to represent f (x, y), the first is to use two axes to determine the spatial location, the third coordinate is the F (Grayscale) value with two spatial variables x and y as functions, and in the second case, the grayscale of each point is proportional to the F value at that point. There are only three equal-spaced grayscale values in the graph. If the grayscale is normalized to the interval [0,1], then the grayscale of each point in the image has a value such as 0,0.5,1, and the third is to simply display the value of F as an array (matrix), it is very difficult to print the entire matrix, and it conveys little information. However, when developing an algorithm, this representation is useful when part of the image is printed as a numeric value.
From the above we can draw a conclusion that the image display allows us to observe the results quickly, the numerical array for processing and algorithm development. A numeric array can be represented by a real matrix, where each element in the matrix is called an image unit, an image element, or a pixel.
Notice that the origin of the digital image is in the upper-left corner, where the positive x-axis extends downward and the positive y-axis extends to the right. Sampling processing can be seen as the process of dividing the XY plane into a grid, at which point the gray value is assigned to a function of each particular coordinate pair, and the process of assignment is the quantization process mentioned earlier.
For storage and quantification of hardware considerations, grayscale progression is usually taken as an integer power of 2. Suppose that discrete gray levels are equal intervals, and they are integers. This book defines the dynamic range of the image system as the ratio of the maximum measurable grayscale to the smallest detectable grayscale. For image contrast, we define the gray scale difference between the highest and lowest gray level ships in an image.
Saturation is the highest value (note that the entire saturated area has a constant high gray level) that the gray level above this value will be cut off.
2.4.3 Space and Grayscale resolution
Spatial resolution can be explained in a number of ways. The most common measure is the number of lines in the unit distance (the amount of line pairs) and the unit distance (pixels). The widely used image resolution is defined as the maximum number of line pairs that can be distinguished within a unit distance. The measurement of spatial resolution must be defined in terms of spatial units to make sense. Grayscale resolution refers to the smallest change that can be distinguished in the gray level.
As the detail in the image increases, the preference curve tends to become more vertical. This result shows that for images with a large amount of detail, there may only be less grayscale levels.
2.4.4 image interpolation
Interpolation has a wide range of applications in tasks such as magnification, shrinkage, rotation, and geometry correction. The use of interpolation in amplification and contraction is a basic method of image resampling.
In essence, interpolation is the process of obtaining unknown positional values with known data. If we need to enlarge an image of 500x500 pixel size by 1.5 times times to 750x750 pixels. A simple amplification method is to create an imaginary 750x750 network that has the same spacing as the original image and then shrinks it so that it accurately matches the original image. Obviously, the pixel interval of the shrunk 750x750 mesh is less than the pixel interval of the original image. In order to give gray values to each point covered, look for the closest pixel in the original image and assign the gray value of the pixel to the new pixels in the 750x750 grid. After the gray-scale assignment of all the points covered in the grid is completed, the image is expanded to the original specified size and the enlarged image is obtained. This method we call the nearest neighbor interpolation. But this approach has the tendency to produce unwanted human defects, such as severe distortions of some straight edges.
A more practical approach is the bilinear interpolation. In this method, we use 4 nearest neighbors to estimate the grayscale of a given position. The order (x, y) represents the position coordinates for which we want to get the grayscale value, so that V (x, y) represents the grayscale value for that position. In bilinear interpolation methods, V (x, y) =ax+by+cxy+d, where 4 unknown coefficients can be determined by the 4 nearest neighbor points (x, y). bilinear interpolation can give a much better result than in the nearest neighbor. The next high degree of complexity is a double three interpolation. Typically, a dual three-time interpolation is better than bilinear interpolation in maintaining detail. Dual three-time interpolation is a standard interpolation method for commercial image editing programs such as Adobe Photoshop and Corel PhotoPaint.
In interpolation, it is possible to adopt more neighboring points and more complex techniques. For example, using splines and wavelets, in some cases, you can get a better result than the above method. For three-dimensional graphics and medical image processing, preserving fine details is particularly important, and for ordinary digital image processing, less consideration of the additional computational burden is reasonable, so bilinear interpolation and double three interpolation is a typical method chosen by people.
