Communication Principles in Image Processing - Gonzalez's Reading Notes (4)

overBeauty reconstruction requires the use of infinite sums to interpolate, and in practice, some kind of approximation has to be sought. In image processing, the most common application of 2D interpolation is to resize an image. Zooming in can be thought of as oversampling, and zooming out is undersampling. Whether it is oversampling or undersampling, interpolation is required, and there are nearest neighbor, bilinear, and bicubic interpolation. In the process of image registration to build an image pyramid, and the last step of image perspective transformation of image registration, interpolation is required, and the quality of the interpolation directly affects the final effect.

A special case of nearest-neighbor interpolation associated with oversampling when we want to increase the size of an image by an integer multiple is the upscaling operation with pixel copying. For example, to upscale an image by a factor of two, we can Copy each column first, then double the horizontal size, and then copy each row, double the vertical size. Undersampling can be achieved by row-column deletion. For example, when shrinking to half, delete every other row and column. When shrinking to non-integer, use grid simulation method, because the essence of interpolation is to use known data. The process of estimating the value of the unknown location. Blurring the frequency domain before downscaling the image reduces aliasing. When the edge content of the image is strong, the effect of aliasing appears to be a blocky image component known as aliasing.

Another artifact is the moiré pattern, which results from sampling the scene with periodic or near-periodic components. In optics, a moiré pattern refers to a beat pattern created between two approximately equally spaced gratings. In digital image processing, this problem usually arises when scanning media prints such as newspapers and magazines, or in images with periodic components, if the interval is comparable to the sampling interval. Moiré patterns are more common than artifacts caused by sampling. Superimposing one pattern on another is mathematically equivalent to multiplying two patterns.

Printed materials such as newspapers just mentioned use halftone diao dots.

Knowing the author : Monster Matt : Continuous tone (continuous-tone) refers to the basic elements (pixels) of the image that contain continuous changes in the depth and brightness of the color, while the basic elements of the half-tone (half-tone) image itself ( Here it refers to the exposure point smaller than the halftone dot) There are only two states of colored and colorless, the printed matter changes from light to dark, and from light to dark is realized by the size of the exposure point. "Using uniform arrangement of exposure points to form dots (AM screening)", "Using uneven arrangement of exposure points to form images (FM screening)" or a combination of the above two methods (second-order FM screening) to achieve image reproduction . The larger the dot area, the darker the color, which is called dark tone; the smaller the dot area, the lighter the color, which is called bright tone. Because the dots have a certain distance in space, they are released discretely, and because the number of screening series is always limited, it cannot achieve infinite changes at the level of the image like a continuous image, so the screened image is called halftone. image.




Guess you like

Origin http://43.154.161.224:23101/article/api/json?id=325682331&siteId=291194637