programmable logic device
A. V. Orlov, S. V. Degtyarev
The image filtration is one of fundamental operations of computer sight, image recognition and image processing. Operation of the overwhelming majority of image processing methods begins with filtering of original images.
The image filtration provides suppression (or smoothing, depending on filter type) pulse noise, and also smoothing of differences of brightness of edges that reduces probability of ruptures occurrence. Pulse noise can appear as a result of imposing of hindrances in the data link or at errors of numbering of images from analogue carriers. Also on the image there can be an additive low-frequency noise.
As a filtration of images understand the operation, having the result the image of the same size received from initial on some rules. Usually intensity (colour) of each pixel of filtering images is caused intensities (colours) of the pixels located in its some vicinity in the initial image. Filtration rules can be the various.
The filtration of digital images is carried out by a "sliding" window of the different size (most used 3×3, 5×5 and 7×7). The essence of the given method consists that some limited two-dimensional area – «the processing window» consistently occupies all possible positions in an image plane. Within a sliding window new value of brightness is calculated as linear or nonlinear function of brightness of the neighbouring points. Accordingly, filters are subdivided into two classes: linear and nonlinear.
Many algorithms of fast convolution for a filtration of images are known, but because of complexity of data processing the problem of fast performance not always dares. Hardware realisation of digital filters which demands use of difficult devices and the considerable hardware resources is one of directions for improve the performance.
There are following architecture of digital filters on programmable logic device (PLD): parallel, consecutive and series-parallel. Digital filters with parallel architecture have the maximum efficiency and the minimum delay, but occupy many logic resources of a microcircuit. Consecutive filters have the least efficiency and the maximum delay, but despite it, more compact in comparison with devices with parallel architecture. In article realisation of the filter with series-parallel architecture is de-scribed.
All calculations made by the presented block of a image filtration, are reduced to operations of multiplication and addition which are carried out using shift registers, adders and logic elements, that allows to realise the considered block on PLD, and operation are carried out in parallel. The result value of brightness of a point on the filtered image turns out on the basis of brightness of points in vicinity 5×5.
Such realisation allows to improve the performance essentially, through usage of high-speed elements and through increase of parallelism, and can have wide use in devices of images preprocessing of the big data files.