A Novel Cross-Hexagon Search Algorithm Based on ... - IEEE Xplore

IEEE International Symposium on Industrial Electronics (ISlE 2009) Seoul Olympic Parktel, Seoul, Korea July 5-8, 2009

A Novel Cross-Hexagon Search Algorithm Based on Motion Vector Field Prediction Shiping Zhu", Jun Tian", Xiaodong Shena, Kamel Belloulata'' "Department of Measurement Control and Information Technology, School of Instrumentation Science and Optoelectronics Engineering, Beihang University, Beijing 100191, China; "Departement d'Electronique, Faculte des Sciences de l'Ingenieur, Universite Djilali Liabes de Sidi Bel Abbes, Sidi Bel Abbes, 22000, Algerie

Abstract-Search patterns and the center-biased characteristics of motion vector distribution have large impact on both searching speed and quality of block motion estimation. In this paper, we propose a new cross-hexagon search algorithm (NCHEXS) using two cross-shaped search patterns as the first two initial steps and large/small hexagon-shaped patterns as the subsequent steps for fast block motion estimation (BME). NCHEXS employs halfway stop technique to achieve significant speedup on sequences with stationary and quasi-stationary blocks. To further reduce computational complexity, NCHEXS employs Modified Partial Distortion Criterion (MPDC). Experimental results indicate that the improvements of NCHEXS over Hexagon Search (HEXS) and Cross-Diamond Search (CDS) can save 45% and 28% of search points while keep similar PSNR, and NCHEXS provides faster searching speed and smaller distortion than other popular fast block-matching motion estimation algorithms. Index Terms-motion estimation, block-matching, hexagon search, modified partial distortion criterion

I.

cross-

INTRODUCTION

Block-matching motion estimation is a vital process for many motion-compensated and video coding standard. Motion estimation could be very computational intensive and can consume up to 60%-80% of computational power of the encode process. So research on efficient and fast motion estimation algorithm is significant. Block-matching algorithms (BMA) are used widely because they are simple and easy to be applied. In the last two decades, many block-matching algorithms are proposed for alleviating the heavy computations consumed by the brute-force full search algorithm (FS) which has the best prediction accuracy, such as the three-step search (TSS), the new three-step search (NTSS) [1], the block-based gradient descent search (BBGGS) [2], the diamond search (DS) [3], the cross-diamond search(CDS) [4], etc. In real world video sequences, more than 80% of the blocks can be regarded as stationary (MV=(O,O)) or quasi-stationary (MV=(±l,O) or (0, ±1)) blocks and most of the motion vectors are enclosed in the central 5 x 5 pixels area for search window [5]. TSS, NTSS and BBGDS employ rectangular search patterns of different sizes to fit the center-biased motion vector distribution characteristics [6]. Hexagon-based search employs a hexagon-shaped pattern and results in fewer search points with similar distortion [7]. In this paper, a novel fast blockmatching algorithm calls cross-hexagon search algorithm

(NCHEXS) is proposed. It uses small cross-shaped search patterns in the first two steps before the hexagon-based search and the proposed halfway stop technique [4]. It results in higher motion estimation speed on searching stationary and quasi-stationary blocks. The traditional algorithms use all the pixels of the block to calculate the distortions that results in heavy computations [8]. We propose a Modified Partial Distortion Criterion (MPDC) that uses certain pixels of the block, which alleviates the computations and has the similar distortion. II.

NEW CORSS-HEXAGON SEARCH ALGORITHM

A.

Cross-center Biased Motion VectorDistribution From observing the motion vector probabilities of different video sequences, we find that most real world video sequences have the center biased MV distribution characteristics. Motionvector probability (MVP) can be concluded as follow: (1) Global optimal distribution is the square-center-biased (SCB) within ± 2 pixels, especially the zero motion vector (ZMV)(O,O); (2) MVP usually decreases away from ZMV; (3) Optional MV found along the vertical and horizontal directions are often more than the other locations with the same radius, which is regarded as cross-center-biased (CCB) MVP distribution. For example, there are about 15.71% and 7.94% of MV found in vertical and horizontal directions with radius of 1 pixel away from the ZMV. This probability is much higher than in the diagonal positions, which totally contribute about 2.76% at the same radius. The results also show that the cross-center MV distribution is more dominant within this radius. For instance, 71.85% of MV are found located in the central 2X2 pixels area, and there is about 68.98% of MV located in the cross-center area. In the 4X4 pixels area, total MVP is 81.75% and the cross-center probabilities within this area is 74.71 %. Due to such a highly cross-biased distribution, the search pattern of BMA should match the cross-center shape to minimize the number of search point while maintain a similar distortion error. B.

New Cross-Hexagon Search (NCHEXS) Algorithm The new cross-hexagon search consists of two patterns: cross-based and hexagon-based patterns. As the motion vectors distribution possesses cross-center biased characteristics

978-1-4244-4349-9/09/$25.00 ©2009 IEEE 1870

(74.76%) in the central 4 X 4 pixels area, two cross-shaped patterns: smalI-cross-shaped (SCSP) and large-cross-shaped pattern (LCSP), as shown in Fig. lea), are proposed as the first two initial steps to the hexagon-based search. There are two different sizes of hexagon search patterns: large and smalI hexagon patterns. The large hexagon pattern used in this paper consists of not only the 7 check points in classic large hexagon pattern, but also the two edge points (up and down), as shown in Fig. I (b). Therefore, the new large hexagon pattern consists of 9 search points which realizes a distinct search speed gain without increasing computational complexity of large hexagon search algorithm as shown in Fig. l(c).

(radius=±2) are checked. The step tries to guide the possible correct direction for the HEXS. Step 4 (LHEXSP): A new LHEXSP is formed by repositioning the minimum MBO found in the previous step as the center of LHEXSP. If the new minimum MBO point is still at the center of the newly formed LHEXSP, then go to step 5; otherwise this step is repeated again. Step 5 (SHEXSP): Switch the search pattern from the large size of hexagon to the smalI size of hexagon (SHEXSP). The four points covered by the smalI hexagon are evaluated to compare with the current MBO point. The new MBO point is the final solution of motion vector. A typical example is shown in Fig. 2. -2

-I

0

+1

-2

+2

0

-I

+1

+2

r'" /

- JJI1"'I r=--

\

1~I

\

.-'-

-I

/

-r

--(I

(b)

/ /

\

\

+1

1

+2

+2

(a)

\ I

0

I

+1

(a)

-I

1

o-

.-1

l' fOJ

-2

-2

\

-2

-1

0

(b)

+1

+2

.J

+3

-2

-I

0

+ 1 +2

+3

+4

/

-J

-I

(c)

o

Fig. I. Search patterns used in the propos ed NCI-IEXS algorithm. (a) 0 LCSP • SCSP O (b) 0 Classic LHEXSP 0 Classic SHEXSP (c) 0 LHEXSP in this paper 0 SHEXSP in this paper

+1 +2

-3

,

-2

--t.

HFft'~ r1

41 - -

-I

+1

2

4

+2

)

+3

I From the simulation results on video sequences, we found that nearly 70% blocks that can be regarded as stationary or quasi-stationary blocks. By having this highly cross-biased property in most of the real world sequences, we take the smalI cross-shaped patterns as the first two steps ofNCHEXS, which will save the number of search points for stationary or quasistationary blocks. Then, search the rest points of LCSP and SCB which is guiding a much more precise direction for the subsequent HEXS. The NCHEXS algorithm is summarized as folIows: Step I (SCSP): A minimum block difference (MBO) is found from the 5 search points of the SCSP located at the center of search window. If the minimum MBO point occurs at the center of the SCSP , the search stops. Otherwise go to step 2. Step 2 (SCSP): A new SCSP is formatted by using the vertex in the first SCSP as the center. If the minimum MBO point occurs at the center of this SCSP , the search stops. Otherwise go to step 3. Step 3 (LCSP): The three unchecked outermost search points of the LCSP, and the two unchecked points of the SCB

4

-2

(c)

(d) -3

-2

-t

0

+ I +2

+4

+3

-4

·3 ·2

41-

-I

+1 +2 +3

(e)

CD First step

@ Seco nd step

& T hi rd step

GJ Fourth step Wifth step

Fig. 2 A search exampl e of the NCI-IEXS algorithm

C. Analysis ofNCHEXS Compared with the current HEXS and cross-diamond algorithm, the characteristic of NCHEXS algorithm lies in reducing the number of search point and increasing search speed, especialIy for (quasi-) stationary blocks(IMVI=I). For stationary blocks, HEXS and current cross-diamond algorithm takes 13 and 9 search points respectively, while NCHEXS takes 5 search points. For quasi-stationary blocks, HEXS and

1871

current cross-diamond algorithm takes 13 and 11 search points respectively, while NCHEXS takes 7 search points.

calculation of partial distortion SAD, (m , n; p , q) is defined as following,

09,(9)

k

• • • • •

I

•

•

SADk(m,n;p,q) = Lsadi(m,n;p,q) i =l

(3) The calculation sequence of sadk(m,n;p,q) (k=l, 2, ... ,16) is as the numbers in Fig. 3, and the pixe ls have an equal distribution in the block. If SADk (m, n; p, q) uses too fewer pixels, it will not correctly replace the SAD. Large simulations find that the percentage of miscarriage of justice will be below 5% if k;:::: 3.

•

• • • • •

•

•

•

III.

EXPERIMETNTAL RESULTS

The proposed NCHEXS algorithm is simulated using the popular video sequences (the first 70 frames of each sequence) . Fig. 3. The calculation sequence of sad, (m , n; p, q) The block size is 16x 16 pixels, and the maximum displacement in the search areas is ±7 pixels in both horizontal and vertical directions. The experiment is proceeded in a PC (CPU: Inter D. Modified Partial Distortion Criterion (MPDC) Core2 E6300, 1.86GHz, RAM: 2G, DDR2). The algorithm is BMA usually uses all of the pixels in the block to calculate implemented in Visual C++ 6.0. the distortion that causes a heavy computation. In fact, we use The average search time per frame by the Total Distortion some parts of the pixels in the block that can cause the similar Criterion and by the Modified Partial Distortion Criterion results. (MPDC) are summarized in Table 1 for each sequence. The block size is 16x 16 pixels, the top left comer Average PSNR values and search points numbers are coordinates of the blocks in n frame and n-I frame are (m,n) summarized in Table 2 and Table 3 for different algorithms and (m + p , n + q) respectively . The sum absolute difference including NTSS, DS, CDS, HEXS, NCHEXS. (SAD) between the blocks in n frame and n-I frame is Table 2 shows that the proposed NCHEXS algorithm always 15 15 SAD(m, n;p,q) = LLJr,,(m+i,n+ j)- J,,_,(m+ p+i,n+q+ (1) consumes the smallest number of search points compared to i =O j =O other fast SMA. The average search points per block with the 1" (m + i, n + j) is the grayscale of (m + i, n + j) point in n observations, NCHEXS