Fast Predictive Coding Method Based on an

3 downloads 0 Views 465KB Size Report
Color System Conversion (RGB to YCbCr): In this stage the RGB color image data is converted to YCbCr color space data. 2. Forward Wavelet Transformation: ...
Fast Predictive Coding Method Based on an Enhanced Block’s Motion Estimation Loay E. George Dept. of computer, College of science, University of Baghdad, Iraq

Aree A. Mhammad Dept. of computer, College of science, University of Sulaimani, Kurdistan region, Iraq

Abstract - An effective and efficient method is proposed for motion estimation in a video sequences. The paper outlines a search technique based on the traditional once-at-a-time search (OTS) and propose a fast predictive version called modified once-at-a-time search (MOTS) algorithms. The modification is based on utilizing the existing inter-block correlation between the motion vectors of the adjacent blocks. The performance of the enhanced modified technique is compared with the modified one in terms of objective quality measurement (PSNR) and searching time. Test results indicate that the performance of the inter frame coding based on using modified block matching algorithm in addition to encoding the error of the highly distorted blocks is better than the results of modified OTS. The results show encouraging reduction of prediction.

To estimate the motion on a block-by-block basis by MOTS requires extensive computations. Several efficient techniques such as the 2-D logarithmic search [6], three-step directed search [7], and a modified version of the former [8] which reduces the complexity, have been developed. The objective of this paper is to develop a much more efficient BMA, called the modified one-at-a-time search (MOTS).These motion estimation techniques are applied to video sequences, and their superior performance compared to the existing techniques is illustrated based on quantitative measures of the prediction error. In Sec-2, the modified OTS and its implementation are presented. In Sec 3 simulation tests evaluating the proposed method are presented. Finally the main conclusions are summarized in Sec- 4.

Keywords: Inter frame motion estimation wavelet predictive shift coding fidelity measure (PSNR) searching time

1

Introduction

Block-based Motion Estimation and Compensation (ME/MC) is a popular inter-frame coding technique used in video compression systems, [1]-[4], to exploit similarities between adjacent frames in typical video sequences. It provides additional compression compared to simple intra-coding methods and also improves fidelity of decoded video in low bitrate applications. In a typical application, the frame being coded is partitioned into fixed sizes blocks and only motion information of these blocks, with respect to a reference frame, is coded. This motion information consists of motion vectors and prediction errors for each individual block. The former estimates the motion on a pel-by-pel basis, whereas the latter predicts the motion on a block-by-block basis. The latter, although computationally simpler, assumes that all the pels within the block have a uniform motion [5]. The former, on the other hand, is more representative of the real motion but at the cost of complex arithmetic.

2

Modified OTS Algorithm

OTS is a simple technique in a two-variable optimization to adjust one variable at a time for minimum distortion [9]. The matching criteria used in this technique are the mean absolute difference MAD.

MAD = ∑ y

∑ Ir ( x + dx, y + dy ) − Ic( x, y)

(1)

x

Where Ir and Ic are reference and current frame. MOTS uses the predetermined motion vectors of the previous blocks (i.e., top, left, and top block to calculate the initial motion vector of tested blocks). The main advantage of this modification is determining motion vectors with minimum prediction error, which is lower than that obtained by using traditional OTS. The initial motion vectors (dx, dy) used in MOTS method are calculated according to the following

1 [ x (i − 1, j − 1) + x (i , j − 1) + x (i − 1, j )] 3 1 ∆ y = [ y (i − 1, j − 1) + y (i , j − 1) + y (i − 1, j )] 3 ∆x =

( 2) ( 3)

equations. 2.1

Modified OTS Algorithm

Figure 1 illustrates the flowchart of the MOTS method.

using a simplified intra frame encoder. To encode such blocks, the wavelet transform is first applied and then followed by hierarchical quantization, and finally the shift coding is applied on the whole block size level. Due to the high distortion which may occur in some blocks of the frame, the MOTS method has not yielded good PSNR results. However, the effect of this distortion was reduced when the highly distorted blocks have been coded using wavelet intra frame coding scheme. Figure (2) shows the flowchart of enhanced MOTS when the stage of checking the distortion level of the motion compensated blocks is added. In this scheme some additional steps belonging to a simplified intra coding method (like color conversion, wavelet transform, hierarchy quantization, and shift coding) are used.

Fig (1) MOTS algorithm’s flowchart 2.2

Enhanced MOTS Method In MOTS method, all blocks of the coded frame are passed through the block matching process for estimating its motion vector relative to previous frames. In some block matching instances, it was found that the best matched block has relatively high distortion level (MAD > Threshold), so such blocks should be coded

Fig (2) Enhanced MOTS algorithm’s flowchart

2.3

Distorted Blocks Coding

As explained in section 2.2, the highly distorted blocks are coded by performing the simplified intra frame coding (see Figure 3) as illustrated in the following diagram:

2.4

Motion Compensation

In video compression, Motion compensation describes a picture in terms of where each section of that picture came from, in a previous picture. Subsequent frames are very similar, thus containing a lot of redundancy. Removing this redundancy helps achieve the goal of better compression ratios. A first approach would be to simply subtract a reference frame from a given frame. The difference is then called residual and usually contains less energy (or information) than the original frame. The residual can be encoded at a lower bit-rate with the same quality. The decoder can reconstruct the original frame by adding the reference frame again. Figure 4 illustrates the compensated frame which contains of reference image with the motion vectors using modified OTS and enhanced MOTS block matching algorithm.

Fig (3) The diagram of simplified intra coding The steps of intra frame coding module. It consists of four stages, they are: 1.

Color System Conversion (RGB to YCbCr): In this stage the RGB color image data is converted to YCbCr color space data.

2.

Forward Wavelet Transformation: In this stage the data of Y,Cb,Cr color bands are transformed from spatial domain to scale-shift domain.

3.

Hierarchical Quantization: This stage receives the wavelet coefficients (float numbers) from the previous stage and maps them to integers.

4.

Entropy Coding: In this stage, the quantization indices are further compressed using entropy coding and represent the codewords results as a stream of bits.

Fig (4) Top: reference frame Mid and Bottom: Compensated frame using MOTS and enhanced MOTS

3

Results and Performance

The modified OTS algorithm is compared with traditional OTS algorithm in terms of quality measurement (PSNR) and searching time. In Figure 5 the variation of PSNR is shown versus number of frames.

Figure 7 shows the variation of PSNR versus frame number (same sequence used in Fig 5) when using enhanced MOTS and OTS method with block size 8x8. The block size is chosen based on best quality measure by testing different block sizes. The search time has also been calculated. Without using intra frame coding to handle the highly distorted blocks, the obtained PSNR value is around 27 dB for all tested frames. The obtained results after using intra frame coding have raised to 28-29 dB.

Fig (5) Comparison between the obtained performance parameters (PSNR and search time) for OTS & MOTS

The performance efficiency comparison was based on PSNR and searching time, as illustrated in Fig 5. The quality of the reconstructed frames coded by MOTS is better than those obtained by OTS (PSNR ~ 27 dB). Also, the time search is considerably reduced by MOTS around (80%). The inter frame coding of the enhanced MOTS and OTS depends on many parameters (i.e., block size, distortion threshold for block’s residual coding, hierarchy quantization). The effects of these parameters have been tested in terms of fidelity measure PSNR and compression ratio C.R. Figure 6 shows PSNR versus C.R when enhanced MOTS method is applied.

16 14 12

C.R

10 8 6 4 2 0 27

27.5

28

28.5 PSNR

. Fig (6) C.R versus PSNR

29

29.5

Fig (7) Comparison between the obtained performance parameters (PSNR and search time) for OTS & enhanced MOTS

Finally, the inter frame coding performance in terms of MSE (Mean Square Error), PSNR, compression ratio and searching time is illustrated in table 1. Table (1) MSE, PSNR, C.F and searching time for different block size using enhanced MOTS and MOTS

4

Conclusion

The performance of the proposed algorithm is based on quality of the reconstructed frames and time searching. The tested result shows that the quality maintains the different values for different block sizes. Test results show that the best block size chosen to obtain a good result was (8x8) pixels, the comparison was based on calculating the fidelity measure PSNR for different video sequences. Test results indicate that performance of MOTS is better than the traditional OTS version, and it showed encouraging reduction in motion estimation time. the inter frame coding becomes more efficient when using MOTS method instead of OTS, and its performance (in terms of PSNR) is improved when the stage of intra frame coding is added to recover the highly distorted blocks. The obtained test results shown that the performance in terms of PSNR is around (29 dB), and the search time is 80% times of the required search time of traditional OTS method. The future works will be an improvement for the efficiency of compression using variable block size partitioning method. Smaller blocks are used to cover the image regions which have high details, and larger blocks for regions with lower details.

5

References

[1] B., Zhao, S., Li, S.: The Application of Wavelet Theory in Video Compression. IEEE, International Symposium on Microwave for Wireless Communication. 2, (2005) 1234–1236. [2] Chen, P., Woods, J.: Video Coding for Digital Cinema. Proceedings of the IEEE International Conference on Image Processing. Vol.1, Rochester, NY (2002) 749–752. [3] Luo, L., Li, J., Zhuang, Z.: Motion-Compensated Lifting Wavelet and its Application in Video Coding. In. Proceedings of the IEEE International Conference on Multimedia and Expo., Tokto, Japan (2001) 481484. [4] Flierl, M., Girod, B.: Investigation of MotionCompensated Lifted Wavelet Transforms , In. Proceedings of the Picture Coding Symposium, SaintMalo, France (2003) 59-62.

[5] Rehan, M., Agathoklis, P., Antoniou, A.: A New Motion-Estimation Technique for Efficient Video Compression. IEEE Pacific Rim Conference. 1(1997) 326-329. [6] . Soongsathitanon, S., Dlay, S.: Block-Based Motion Estimation Using a Novel Orthogonal Logarithmic Search Algorithm. IEEE, Visual Information Engineering, 1(2003) 246-249. [7] Koga, K., Iinuma, K, Hirano, A.: MotionCompensated Interframe Coding for Video Conferencing. IEEE Proceedings NTC (1981) 531534. [8] Reoxiang L., Bing Z., Liou, M.: A new three-step search algorithm for block motion estimation. IEEE Transaction on Systems for Video Technology (1994) 438-442. [9] Srinivasan, R., Rao, R.: Predictive Coding Based on Efficient Motion Estimation. IEEE Transactions on Communications Vol. COM-33 (1985) 888-896.

Suggest Documents