RECURSPIVE SEPARABLE 2D DIGITAL FILTER FOR INCREASING THE SHARPNESS OF RGB IMAGES
К.О. Sever, D.A. Guzhva, Игорь Ильич Турулин · Известия Южного федерального университета. Технические науки · 2023
An important role in the perception of image quality is sharpness, that is, the An importantrole in the perception of image quality is played by sharpness, that is, the magnitude of the brightnessgradient in areas near the boundaries of objects. This characteristic is responsible for theclarity and detail of small image elements. Defocusing the camera lens and insufficient illuminationare the main factors that can lead to digital image blurring. To increase the sharpness, variousprocessing methods are used, such as filtering in the frequency domain, for example, the use offast Fourier transform to emphasize the boundaries and textures of the image. The use of this typeof filtering allows you to control the contrast and frequency content of the image, which leads toan improvement in visual perception. However, this method has a number of significant drawbacks,such as logarithmic complexity and performing additional calculations associated withforward and inverse Fourier transforms. Therefore, the preferred method of image sharpening isthe so-called spatial processing, which provides direct filtering of image pixels without additionaltransformations, and the reuse of processing results (recursive component) in the filter allows youto reduce the number of operations, reduce computational complexity. The article describes thedevelopment of an effective recursive separable two-dimensional digital filter to sharpen largedimensionalRGB images. The algorithms of its construction are given, the corresponding blockdiagrams are designed. The filter has the property of more uniform detail of image objects, and isless susceptible to the creation of pulse noise. Also, for the original high-resolution RGB image, ablur filter is modeled, the matrix of which is filled according to the normal (Gaussian) law.To assess the filtration quality, the developed filter is compared with the algorithm of classicaltwo-dimensional convolution with a 5x5 Laplace high-pass filter core.