Paper 106: An Introduction to Image enhancement

Image enhancement is a pixel-to-pixel operation that changes the grayscale histogram of an image in a predetermined manner. Sometimes referred to as contrast enhancement, gray-level transformations. The point operation is unlikely to change the spatial relationship within the image, and the grayscale value of the output pixel is determined by the value of the input pixel. Its role:

Contrast enhancement: Expands the contrast of the feature of interest.

Photometric calibration: Removes the nonlinear effects of image sensors. The so-called photometric, is in the visible band, taking into account the subjective factors of the human eye after the corresponding metrology subject called photometric. Calibration calculates the internal reference and calculates the internal reference in the photometric category.

Display calibration: Overcomes the non-linear factor in the image preservation and display process.

Contour Line: thresholding

Cropping: The output grayscale is cropped to between 0~255.

First, explain what the contrast is:

Popular speaking, is the contrast degree of light and shade, usually shows the clarity of the image quality, calculation formula: For example: Set the image L = {1 3 9 9 2 1 3 7 3 6 0 6} c = [(1^2 + 2^2) + (2^2 + 6^2 + 2^2) + (... ) + ...  ]/48 = 16.6818. In fact, the contrast reflects the light and dark areas in an image of the brightest white and the darkest of the dark between the different levels of brightness of the measurement, that is, the size of an image grayscale contrast, the larger the difference range represents the greater the contrast, the smaller the difference, the better the contrast ratio of 120:1 can easily display vivid, rich color, When the contrast rate is as high as 300:1, the color of each order can be supported. But the difficulty of having the same ratio and brightness is not yet an effective and fair measure of the ratio, so the best way to identify it depends on the user's eye.

The purpose of grayscale transformation (image enhancement) is to improve the picture quality, so that the image display effect is more clear. Mainly includes:

Linear contrast broadening

The aim of contrast broadening is to emphasize the parts of people's concern. The principle is: pixel-level point-to-point grayscale mapping. As follows:

F (i,j) represents the original image, G (I,J) represents the processed diagram, requiring g,f to change in the [0,255] interval, and G is better than F. The mapping relationship is calculated as: in the [0,a] interval, g (i, J) =α* F (i,j);In the [A, b] interval, g (i,j) =β* (f (i,j)-a) + GA; in the [B, 255) interval, g (i,j) =γ* (f (i,j)-B) + GB; Above, α,β,γ are slope. This is called contrast broadening. At this point, if we only display the specified interval of G, i.e. α,γ is set to 0.G (I,J) =β1 * (f (i,j)-a); The above mapping becomes: This process is called grayscale window slicing, to get gray-level windows. That is, only the information that is within the specified gray level range is displayed. Contrast-enhanced targets with grayscale values falling within a certain range. We will need to detect the target and other parts of the screen to separate, the target part is set to 0/255, the non-target portion is set to 255/0, that is, the process of α,γ 0, which is called grayscale window slicing. The effect of contrast broadening: Dynamic Range AdjustmentDynamic range, refers to the image recorded in the scene from the dark to bright changes in the range, because the human eye can be distinguished by the range of gray scale is limited, so when the dynamic range is too large, the high brightness value of the dark area of the signal is masked. Pass compression of dynamic range, you can extend the range of changes in the gray level of the part you care about. For dynamic range adjustment is divided into: 1. Linear dynamic range adjustment. This process is the above contrast broadening, through the gray window slicing, to get the process of gray-level windows. At this point, the calculation formula in the gray window is: g = 255/(b-a) * (f (i,j)-a); Effects of linear dynamic range adjustment: 2. Nonlinear dynamic Range Adjustment: Because the human eye to the signal processing is a link of the approximate logarithm operator, therefore, the method of logarithm is usually used.    G (I,J) = c * LOG10 (1 + f (i,j)), where C represents the maximum amount of gray you need to suppress. For example, we need to extend the dark part and suppress the bright part, at which point g (i,j) = 255 * log (f (i,j) + 1); Other methods for nonlinear dynamic range adjustment: The classification is based on the gray level of the intermediate range. 1. Increase the mid-range grayscale and only make minor changes to the dark/bright pixels. Formula: F (x) = x + Cx (dm-x), where the Dm is the maximum value of the grayscale level, C defines an increase in the median gray range (c> 0) or a decrease (C < 0) to a demo:2. Reduce the contrast of lighter or darker objects to enhance the contrast of objects in the middle of the gray level, medium The slope is greater than 1, and the slope of the two ends is less than 1. α: (0,1);

corresponding grayscale:

3. Depress the contrast in the middle gray level, while in the lighter or darker parts of the contrast enhancement, the median slope is less than 1, the ends are greater than 1.

Curve in contrast to Class 2, the 2nd and 3rd classes are given a demo: Nonlinear dynamic Range adjustment effects: Histogram equalization Processing the idea of histogram equalization is to broaden the number of pixels in the image, and reduce the number of pixels in gray scale, so as to achieve the purpose of clear image. Its mathematical principle is actually from a distribution (the input luminance histogram) is mapped to another distribution (a wider, ideal uniform luminance value distribution), that is, the mapping function should be a cumulative distribution. You can use the cumulative distribution function to map the original distribution as a balanced stretch distribution. principle:The point operation allows an input image to be converted to an output image with the same pixel points at each gray level (the output histogram is flat), which is useful for converting an image to a consistent format before image segmentation or segmentation. After equalization, the number of pixels per level is A0/DM,DM is the maximum grayscale value of gray level, A0 is the area of the image. Equalization Method: 1. Find the histogram of the original image, H = {x0,... xi}i in the 0~255 range. 2. Calculate the probability of gray distribution, the total pixel of the image n = L * W, the number of pixels per gray level in the entire image as a percentage. HS (i) = h (i)/n (i in 0~255). 3. Calculate the cumulative distribution of the gray level Hp:4. Calculates the grayscale value of the new image. Histogram equalization Effect: pseudo-color technology because the ability of the human eye to recognize different colors is stronger than the ability to distinguish different gray levels, it is the basic basis for pseudo-color enhancement that the gray-scale changes that cannot be distinguished by human eyes are applied to different colors to improve the recognition rate. In view of the gray image to generate a color image is a one to three mapping, obviously less information to obtain more information is necessarily based on the estimation principle, that is, the unknown part, through various means of reasonable estimation, the purpose of the study , the estimated methods are different, and there are several ways to do this:1. A pseudo-color method based on luminance representation: According to the description of the temperature, when the temperature is low, we will think of blue, also known as cold color, when the temperature is higher, we will think of red, also known as warm color, according to the characteristics of the human senses, the low brightness map to blue, high brightness map to red. As a result, pseudo-color processing can be performed as follows: 2. Pseudo-color method based on region representation: This pseudo-color technique is preprocessing the original image, which will recognize the different colors of different areas. Mainly used in medical diagnosis, the aim is to highlight the lesions.

Paper 106: An Introduction to Image enhancement