A Method of Interlaced Video Conversion Based on Pattern Recognition

Abstract

The technical scheme of the invention comprises an interlaced video conversion method based on pattern recognition. the method includes: Training samples by offline, artificially specifying the most appropriate interpolation modes in a plurality of typical scenes and calibrating samples; then, extracting multi-dimensional features of the interpolation points; performing Supervised training, obtaining a feature mapping matrix according to an interpolation mode, extracting multi-dimensional features of a current interpolation point during online interlacing data conversion, obtaining an interpolation type most suitable for the current interpolation point through a mode recognition method, selecting corresponding probabilities and interpolation calculation results thereof according to variousinterpolation modes, mixing the probabilities and the interpolation calculation results, and outputting a final video signal. The beneficial effects of the invention are as follows: the paint has good effects; The mode recognition method is used for replacing manual simple logic to classify the interpolation scene, the classification accuracy is improved, the image quality is improved, meanwhile,the mode recognition method has higher adaptability to various video scenes, the used interpolation implementation mode can be reasonably selected according to the cost, and the implementation cost and effect of the conversion algorithm are balanced.

Description

technical field [0001] The invention relates to an interlaced video conversion method based on pattern recognition, which belongs to the field of video signal processing. Background technique [0002] Interlaced video is a video storage and transmission format that downsamples images in time and space to save transmission bandwidth and maintain image quality as much as possible. It is widely used in television and monitoring fields. Until today's era when digital video transmission methods are popular, there are still wide applications. For the current mainstream progressive display devices, due to their inherent characteristics of time-space interlaced sampling, tearing and jagged phenomena appear in the display of interlaced video, which requires special conversion devices. To solve the display problem of interlaced video on progressive display devices. [0003] The device for converting interlaced video to progressive video generally includes the following processes: ...

AI Technical Summary

Problems solved by technology

[0010] First of all, because of the complexity of video images, different interpolation methods are often suitable for different scenes, and there will always be interpolation errors or unreasonable phenomena in different areas of different images. There are many papers in the industry to study various algorithms in various scenarios For example, linear filtering, such as vertical filtering, has a better effect in static flat areas or vertical edges, but the effect is relatively poor in areas with large motion ranges or rich textures; edge-based interpolation methods at large angles, Areas with large motions are better, and interpolation on small-angle edges is prone to errors; while the motion compensation method based on block matching provides better results for areas with small motions, the search and matching effect for large motions is not good and high cost

[0011] Secondly, the traditional algorithm extracts a small number of features in real time online for simple judgment, and the shortcomings are also obvious. A small number of features are often not enough to describe the characteristics of the current interpolation point

In fact, the selection of the interpolation result type is actually a determination of the interrelationship of multi-dimensional factors, and it is often not easy to summarize the regularity of simple description through human observation

The usual practice is to start from one or two simple scenes, manually try to limit the range of several parameters and the simple relationship between each parameter to determine which interpolation algorithm to choose under which conditions, and then continue to add conditions to adapt to various videos scenes, it takes more time and manpower, and it is not easy to achieve a better balance between various scenes

The fundamental reason is that it is difficult to generalize the complex relationship between multi-dimensional factors through human observation and use simple (if else) logic to judge

Images