Spatial filter Enhancement

The previous period summed up the image noise classification and distribution, and then went out to play for a few days. Now proceed to the previous content to continue the study. Image noise is an important cause of image quality reduction, improve the quality of the image has two kinds of methods, such as image enhancement, that is, the reasons for not considering the quality of the image, only the parts of interest in the image processing or highlighting the useful image features, so the improved image is not necessarily to approximate the original image, mainly to improve the understanding of the image Another kind of method is image restoration technology, which needs to compensate for the quality of the image, and make the improved image approach the original image as much as possible.

Here you will learn image enhancement processing first. Image enhancement method can be divided into spatial domain method and frequency domain method, and space domain method can be divided into point operation (Gray scale transformation) and Neighborhood/template operation (spatial filtering). The histogram processing described in the previous articles is a point operation, so here we mainly study the neighborhood/template operation, that is, the spatial filtering method. (1) Template operations and convolution operations

Spatial filtering is done by using image template for neighborhood operation, and template operation is a commonly used method in digital image processing. As shown below:

The upper style is somewhat similar to a matrix, often referred to as a template. The template operation means that the template is covered in the original image so that the center of the template corresponds to the pixels to be processed, and then the pixels that need to be processed and their eight neighborhood pixels are multiplied by the corresponding pixels of the template, and the resulting value is the new value of the pixel to be processed. The operation is then repeated, so that the template continuously passes through each pixel in the original image, resulting in a new image.

The template operation implements a neighborhood operation where the value of a pixel is not only related to this pixel point, but also to the value of its neighborhood point. In fact, the mathematical meaning of the template operation is the famous convolution operation. Because the same is the process of weighted summation. Templates are equivalent to convolution cores in convolution operations. However, there are some problems with template operations (convolution operations):

1. Image boundary problem. When moving a template to an image boundary, there is no complete neighborhood for the pending pixel, so the pixel values on the four edges of the original image cannot be computed. There are many ways to solve the problem, and it is simpler to ignore image boundary data.

2. Dynamic range problem for new pixel values. The weighted sum is usually greater than the original pixel value, which requires multiplying the sum result by an attenuation factor and finally limiting the result to 0~255 between the results. (2) Spatial filter function Classification

The

Various spatial filters can be divided into smooth and sharp according to the function. Image smoothing tends to blur the boundary and contour in the image, in order to reduce the effect of this kind of unfavorable effect, it is necessary to use image sharpening technology to make the edge of the image become clearer. The aim of image sharpening is to make the image's edges, contour lines and image details become clear, and the fundamental reason for the blurred image is because the image is averaged or integral, so it can be inverse (such as differential operation) to make the image become clearer. Considering the frequency domain, the essence of the image blur is that the high-frequency component is attenuated, so the image can be made clear with a pass filter.