CN1581977A - Tree-structure-based grade tree aggregation-divided video image compression method - Google Patents

Abstract

The method includes steps: first at encoding end, obtaining distribution of image energy on time and frequency domain through discrete wavelet transform; based on correlation between wavelet coefficient, wavelet coefficient in each order is divided according to tree structure; SPIHT encoding is carried out for wavelet coefficient of each tree; encoding result is deposited on encoding end momentarily; finally, results of encoding each tree are synthesized as a code stream in use for storing or transferring. The invention realizes video compression in high compression ratio and low degree of distortion by using less memory space, particular suitable to dedicated system realized through hardware.

Description

Video image compressing method is divided in hierarchical tree set based on tree

Technical field

The invention belongs to the video image compression technology field, be specifically related to a kind of hierarchical tree set and divide video image compressing method based on tree.

Background technology

Hierarchical tree set division (SPIHT) algorithm has taken into full account the correlation between the data, and has considered that also the higher bit data importance is higher than the characteristic of hanging down Bit data in the same data when coding.So use the distortion factor that the SPIHT method is compressed, the decompressed video image can obtain not increasing than higher compression ratio the decompression result, so this method has received increasingly extensive concern.In the process of specific implementation, coded system need be set up three chained lists, promptly important pixel chained list (LSP), inessential pixel chained list (LIP) and inessential pixel set chained list (LIS), and these three chained lists are used for writing down the intermediate data of tree division.By the use of chained list, the sequence arrangement that encoding code stream can descend according to threshold value (importance), thus guarantee the transmission of important information and can block non-important information, blocked the compression effectiveness of the high compression ratio of code stream arbitrarily.In order to improve compression effectiveness, require tree to comprise more data.But along with the increase of data volume in the tree, the length of three chained lists is just more and more longer, just requires huge memory headroom in actual applications, and this has just increased the cost and the complexity of system.So under the prerequisite that does not reduce compression effectiveness, the method that shortens chained list length is just becoming the focus of research.

Summary of the invention

The objective of the invention is to propose a kind of hierarchical tree set and divide (SPIHT) video image compressing method,, save the memory headroom expense of system to guarantee shortening the length of chained list greatly under the prerequisite that compression effectiveness do not descend based on tree.

(SPIHT) video image compressing method is divided in the hierarchical tree set based on tree that the present invention proposes, the concrete steps of coding are as follows: at first by wavelet transform obtain image energy the time distribution on the frequency domain, because the flatness of image, image energy concentrates on low frequency part; According to the correlation between the wavelet coefficient, wavelet coefficient at different levels is divided according to tree; Wavelet coefficient to every tree carries out the SPIHT coding respectively then, and coding result temporarily leaves coding side respectively in; At last the coding result of every tree is synthesized a code stream and be used for storage or transmission.

According to the data dependence between the wavelet coefficient, it is tree root that wavelet coefficient at different levels is divided each coefficient that is meant with the lowest frequency subband according to tree, obtains the data of each point in the tree according to the data dependence of wavelet coefficient position between the different stage.In the tree, the relation between upper level wavelet coefficient and the next stage wavelet coefficient is called father and mother and children or offspring's relation.In wavelet coefficient, the coefficient of different sub-band same position, similitude is often numerically arranged, according to such relation, with each coefficient of lowest frequency subband root node as tree, the coefficient of same position is as the first order children of tree in the higher leveled filial generation, with the coefficient of children's same position of each first order children as first order children, also is the second level children of tree in the more higher leveled filial generation ... up to the coefficient of high-frequency sub-band children as afterbody.

Wavelet coefficient to every tree carries out the SPIHT coding respectively, can reduce simultaneously treated wavelet coefficient, the intermediate object program that produces is less, has shortened the length of important pixel chained list (LSP), inessential pixel chained list (LIP) and inessential pixel set chained list (LIS).Its method is exactly that the coding result of every tree is all obtained successively according to the order that threshold value descends, up to threshold value drop to can satisfy compression and require till.The prediction threshold value decision that the threshold value falling-threshold can be obtained by the minimum threshold or the empirical value of former frame group.The result that the wavelet coefficient of every tree carries out SPIHT coding will not directly be transmitted, but is temporarily stored in coding side, when depositing the encoding code stream under each threshold value situation is deposited successively, and is write down encoding code stream length under each threshold value situation.

After the wavelet coefficient end-of-encode of all trees,, need the coding result of every tree is synthesized the target code stream in order to obtain meeting the target code stream of compression ratio requirement.The method of synthetic code stream is to determine minimum threshold value, be called interceptive value, make the code stream sum that is not less than this threshold value in every tree-encoding code stream be not more than target code stream length, these encoding code streams and code stream length are synthesized the target code stream, and remaining target code stream is again by all the other encoding code stream mean allocation of every tree.The length that exactly threshold value among every tree-encoding result is not less than the code stream of interceptive value and these code streams is directly as the target code stream, the part of target code stream deficiency by threshold value among every tree-encoding result less than the code stream mean allocation of interceptive value.

Depositing of the division that focuses on tree of cataloged procedure, tree wavelet coefficient coding result and synthesizing of target code stream.

In decoding end, decode procedure is the inverse process of cataloged procedure: the buffer memory of at first code stream to be decoded being distributed to every tree, the code stream that every tree is assigned to carries out the SPIHT decoding successively again, obtain the wavelet coefficient of tree, wavelet coefficient with tree is reduced to the wavelet coefficient of arranging by subband again, obtains decoded picture by wavelet inverse transformation.

(SPIHT) video image compressing method is divided in hierarchical tree set based on tree proposed by the invention, has effectively solved the contradiction between image data amount and the chained list length.In order to improve compression effectiveness, the image (frame group) of multiframe can be carried out wavelet transform together, make every tree can comprise abundant wavelet coefficient; Because every tree is encoded respectively, can't cause the excessive lengthening of important pixel chained list (LSP), inessential pixel chained list (LIP) and inessential pixel set chained list (LIS) length.

Description of drawings

Fig. 1 is root node and first three searching relation for children.

Fig. 2 is the relation that back two generation children seek.

Embodiment

Below to the invention in each composition discussed respectively.

1. wavelet transform result's tree is divided

Wavelet transform can use three-dimensional wavelet transform, promptly carries out wavelet transform respectively at line direction, column direction and time orientation.Each coefficient of the lowest frequency of transformation results is as the root node of one tree, and according to relation of plane down, constitutes tree.The size of supposing the lowest frequency coefficient is W _Min* H _Min, W wherein _MinAnd H _MinBe respectively the width and the height of the lowest frequency subband of lowest frequency frame.

1) its children of root node children finding method are:

2) two-dimentional its children of children's finding method are:

3) three-dimensional its children of children's finding method are:

Illustrated in figures 1 and 2, Fig. 1 represents is root node and first three searching relation for children, what Fig. 2 represented is the relation that back two generation children seek, in seven branches of only having drawn among the figure one.

2. every tree wavelet coefficient SPIHT coding result deposits D _{M, N}The expression threshold value is from 2 ^N+1Drop to 2 ^NThe time M the tree threshold value be the coded data of N.L _{M, N}The expression threshold value is from 2 ^N+1Drop to 2 ^NThe time M the tree threshold value be the coded data length of N.The form that the coding result of all trees is deposited is as follows:

Threshold value	2 N	?2 N-1	……	?2 -1
					The 1st tree	D 1，N	?D 1，N-1	……	?D 1，-1
The 2nd tree	D 2，N	?D 2，N-1	……	?D 2，-1
					The 3rd tree	D 3，N	?D 3，N-1	……	?D 3，-1
					M tree	D M，N	?D M，N-1	……	?D M，-1

3. the target code stream is synthetic

If M tree arranged, the target code stream length of requirement is Q.Drop to 2 in threshold value ^PThe time, total code stream length of all trees is

$N_1=\underset{n=1}{\overset{M}{\mathrm{\Σ}}}\underset{m=P}{\overset{N}{\mathrm{\Σ}}}{L}_{n,m},$ Drop to 2 in threshold value ^P-1The time, total code stream length of all trees is $N_2=\underset{n=1}{\overset{M}{\mathrm{\Σ}}}\underset{m=P-1}{\overset{N}{\mathrm{\Σ}}}{L}_{n,m},$ And satisfy N ₁＜Q≤N ₂, so 2 ^PBe interceptive value.Earlier M tree threshold value dropped to 2 ^PThe time all code streams and code stream length as the target code stream, if the code stream length of every tree target approach code stream will represent that this moment, the target code stream was about N with the X bit ₁+ X is again with remaining Q-N ₁The object code levelling of-X all is assigned in M the tree in remaining encoding code stream, is 2 with every tree threshold value just ^P-1Preceding (Q-N ₁-X)/the M code stream is as the target code stream.Specifically in the target code stream, the encoding code stream of each tree is to arrange like this:

The 1st tree

D _{1, N}---D _{1, P}

The 2nd tree D _{2, N}---D _{2, P}

……

M tree

D _{M, N}---D _{M, P}

D _{1, P-1}In the 1st bit; D _{2, P-1}In the 1st bit; D _{M, P-1}In the 1st bit;

D _{1, P-1}In the 2nd bit; D _{2, P-1}In the 2nd bit; D _{M, P-1}In the 2nd bit;

……

Reach requirement up to target code stream length.

The process of decoding is entirely the inverse process of coding.At first code stream to be decoded is distributed to the buffer memory of every tree, the code stream that every tree is assigned to carries out the SPIHT decoding successively again, obtain the wavelet coefficient of tree, the wavelet coefficient with tree is reduced to the wavelet coefficient of arranging by subband again, obtains decoded picture by wavelet inverse transformation.

The result of emulation

Concrete simulated conditions is as follows:

The Y Value Data of Miss American video image group 1-8 two field picture, every two field picture size is 352 * 288.Carry out three grades of 3 d-dem wavelet transformations, again low-frequency frame is carried out the two-stage two-dimensional discrete wavelet conversion, wavelet basis is selected Daubechies9/7 biorthogonal wavelet (line direction and column direction) and Haar small echo (time orientation) for use.Have 99 trees.

Experimental result is as follows:

Original image	8 frames, 352 * 288 Y values			Totally 811008 bytes
						The wavelet transformation result	Each wavelet coefficient is represented with 16bits			Totally 1622016 bytes
Every tree	A frame component is 99 trees			Totally 16384 bytes
						Compression multiple/target code stream length		Index	Before the optimization		After the optimization *
200/4055 byte		LIP chained list length	6642		4433
						LSP chained list length	6084		6459
		LIS chained list length	3521		536
						The PSNR effect	39.3690		39.0733
		150/5406 byte		LIP chained list length	10752					4429
LSP chained list length	7851						6419
				LIS chained list length	4651			536
The PSNR effect	39.9526						39.5110
				100/8110 byte		LIP chained list length		14384		4443
LSP chained list length	12134		6448
						LIS chained list length	5216		536
The PSNR effect	40.7056		40.4405
						50/16220 byte		LIP chained list length	44952		4390
LSP chained list length	22792		6462
				LIS chained list length	13988			536
The PSNR effect	41.4691		41.3303

^*LIP after the optimization, LIP, LIS are length maximum in 99 trees, and every tree-encoding all proceeds to threshold value and reduce to till 8, and the evidence threshold value drops to 8, generally just can satisfy the requirement of compression ratio.

We find by result of upper experiment, though the result of this SPIHT coding method has reduced PSNR (reducing very for a short time), the space that is used to store chained list can reduce greatly.

Claims (5)

1. Video image compressing method is divided in 1 one kinds of hierarchical tree set based on tree, it is characterized in that by wavelet transform obtain image energy the time distribution on the frequency domain, again wavelet coefficient at different levels is divided according to tree, wavelet coefficient to every tree carries out the SPIHT coding respectively then, coding result is temporarily deposited respectively, at last the coding result of every tree is synthesized a code stream and is used for storage or transmission.
2. Video image compressing method is divided in the 2 hierarchical tree set based on tree according to claim 1, the division that it is characterized in that tree is a tree root with each coefficient of lowest frequency subband, obtains the data of each point in the tree according to the data dependence of wavelet coefficient position between the different stage.
3. Video image compressing method is divided in the 3 hierarchical tree set based on tree according to claim 1, it is characterized in that every tree all carries out the SPIHT coding respectively, the coding result of every tree all obtains according to the order that threshold value descends, up to threshold value drop to can satisfy compression and require till.
4. 4 according to claim 1 or 3 described hierarchical tree set division video image compressing methods based on tree, the coding result that it is characterized in that every tree is temporarily stored in coding side earlier, when depositing the encoding code stream under each threshold value situation is deposited successively, and write down encoding code stream length under each threshold value situation.
5. 5 according to claim 1 or 4 described hierarchical tree set division video image compressing methods based on tree, the method that it is characterized in that synthetic code stream is to determine minimum threshold value, make the code stream sum that is not less than this threshold value in every tree-encoding code stream be not more than target code stream length, these encoding code streams and code stream length is synthetic as the target code stream, and remaining target code stream is again by all the other encoding code stream mean allocation of every tree.