CN106791853A - A kind of ROI Lifting Wavelet method for encoding images of view-based access control model memory models - Google Patents

Abstract

The invention belongs to image/video field of information processing, it is ensured that vision sensor is capable of the work of peak efficiency, then its workload is reduced as far as possible and memory data output is reduced.The present invention is proposed based on a kind of cluster by vision memory model, video sequence is divided into background frames and ROI frames, the signal decomposition of Lifting Wavelet is combined with reconstruction calculations form with interesting sets, using the method for optimized truncation embedded encoded (EBCOT).The method can more effectively realize the compression of image, and the compression of images effect for drawing has better conformed to human eye vision engineering.

Description

A kind of ROI Lifting Wavelet method for encoding images of view-based access control model memory models

The invention belongs to video image information process field, be related to the video image background under a kind of complex environment to model, compressed encoding and motion mesh Mark detection.

Background technology

In general coding techniques, it is contemplated that most important of which purpose is the various statistical redundancies letter for needing removal to collect in information Breath, its main foundation is information theory and Digital Signal Processing, but the limit has been arrived in forefathers' research in this respect, and compression ratio is difficult to Improve again.The features such as considering human eye to profile, the specific sensibility at edge and directional perception characteristic, the second generation of having arisen at the historic moment figure As coding method.Study at present it is more it is main have two ways, the first is that the texture and edge of image are respectively processed, The coded system of its treatment is also different, and such case has mainly used the thought of split degree；Second method is firstly the need of right Target carries out the filtering on various directions, obtains the data of different directions, and each data segment can be compiled using specific mode Code, the characteristics of this is mainly in view of anisotropic filtering and human eye orientation-sensitive.Someone takes method of partition and the contracting of self adaptation Various block sort technologies of short scramble time are designed in conjunction, improve Image Coding visual effect, but these methods are list Only entire image is divided into each fixed size block, although what is divided is good again, its compression ratio is difficult to improve.The present invention Propose based on a kind of cluster by vision memory model, video sequence is divided into background frames and region of interest ROI frame, will carry The signal decomposition for rising small echo is combined with reconstruction calculations form with interesting sets, embedded encoded using optimized truncation (EBCOT) method.Because the position of non-omnidirectional's sensor and shooting angle are fixed, the method can be realized more effectively The compression of image.

The content of the invention

Three layers of wavelet decomposition are carried out to image first, corresponding ROI masks are also three layers of decomposition.In order to be carried with small echo below Lifting method is combined, and image is carried out the Parity-decomposition of row and column, is divided into four segments：

F1 1=f ([1：2：M-1], [1：2：N-1])；F12=f ([1：2：M-1], [2：2：N])；F21=f ([2：2：M], [1：2：N-1])；

F22=f ([2：2：M], [2：2：N])；

As needed secondary clearing again, as long as then carrying out four pieces of decomposition to f11 again, by that analogy.In order to save memory space, reduce and calculate Method complexity, Lifting Wavelet uses lazy small echos, and input signal s (i) is divided into 2 groups according to parity, uses even order The predicted value P (s (i-1)) of s (i-1) goes prediction (or interpolation) odd numbered sequences d (i-1).First to traveling line splitting, prediction, renewal and conjunction And, then row are carried out with four operations of the above.By after decomposition, the wavelet coefficient of background image being combined with ROI masks, ROI Part leaves some space, the coefficient after filling three layers of Lifting Wavelet of foreground image.

Choose certain threshold value in the transform domain as illustrated again, filter some small coefficients, reduce memory space.

The characteristics of for ROI, using the embedding space matrix algorithm (EBCOT) of optimized truncation point, by image in mask image Each interested piece is decomposed into, subband is divided first, then area-of-interest block block is individually encoded, produce compressed bit stream. And for background image, due to generation model, without still further encoding.

As long as and it is then that, by anti-renewal, three processes of prediction and merging are to be capable of achieving by image to recover image.

Reconstructed image algorithm based on Lifting Wavelet is simple, it is easy to accomplish, and the template of mask is not needed with regard to original can be recovered Beginning image, effect is preferable.

Claims (2)

1. The present invention puts forward a kind of region of interest ROI Lifting Wavelet method for encoding images of view-based access control model memory models, and the method includes following main points：

According to man memory rule, the present invention propose a kind of cluster by vision memory model based on sorting technique.

2. video sequence is divided into background frames and region of interest ROI frame on the basis of claim (1), the signal decomposition of Lifting Wavelet is combined with reconstruction calculations form with interesting sets, video image is compressed using the method for optimized truncation embedded encoded (EBCOT).