introduction
Motion estimation is to find the pixel (or image block) of the current frame when it is time to find the current frame, and the pixel (or image block) is used as a prediction basis from the pixel (or image block) in this position. The accuracy of the prediction. Since the motion estimation in H.264 uses a series of new technologies, such as seven block sizes (divided into 1616, 168, 816, 88, 8 × 4, 4 × 8, 4 × 4 seven types) Sub block performs motion estimation), 1/4 pixel accuracy motion compensation technology and multi-reference frame technique, etc., the amount of calculation is also greatly increased while compressing at least twice. The experimental results show that the exercise estimates 60% to 80% of the H.264 encoder. The motion estimation in H.264 is composed of an integer motion estimate and a score motion estimate. Since the natural video image sequence or the synthetic video image sequence, the actual subject's motion accuracy is arbitrary, so the introduction of the score motion estimation can accurately describe the movement trajectory of the object, and further remove the time percentage of video image sequences. Yu, its accuracy reaches 1/8 pixel accuracy. The motion vector of the score pixel is shown in Figure 1.
Be
Be
Figure 1 Division Pixel Sport Vector
Generally in practical applications, motion estimates generally use a hierarchical search algorithm: first find the best whole pixel motion vector in the search area, then look for optimal 1/2 matching points under the best matching point of the whole pixel to get half a pixel precision The motion vector, followed by 1/4 pixel point search around the half pixel precision, obtaining 1/4 pixel precision optimal matching point and the corresponding motion vector. Since the quantity of the pixel motion estimation is large, many scholars are optimized from algorithms from algorithms, and propose a lot of fast search algorithms, reducing the number of search points to achieve complexity complexity. This article is based on this purpose. Based on the block matching algorithm, a hardware implementation method of sub-pixel motion estimation of 1/4 pixel accuracy is proposed. On the basis of whole pixel motion estimation, the best matching point search of half pixel precision and 1/4 pixel accuracy is achieved, and there is a higher parallelism on the space, and the hardware is simple and effective.
FME motion vector
Each block or sub-macroblock divided area in the inter-frame encoded macroblock is predicted according to the area of the same size in the reference frame, and the relationship between them is expressed in motion vectors. H.264 When the luminance component and chroma component are sub-pixel search, the motion vector between the two is different, and the brightness component is used in 1/4 pixel accuracy, and the chromaticity component is adopted 1/8 pixel accuracy.
Assuming point h is the best match point found in the entire pixel motion estimation, on which one/2 pixel points are selected, such as points (BB, AA, etc.), if the vertical and horizontal components of the MV are integers, The corresponding pixels actually exist; if one or two are scores, the corresponding brightness and chrominance pixels corresponding to the reference block do not exist, and need to be interpolated adjacent to the encoded point.
The steps to generate within the interpolation pixel are as follows:
First, it is generated to refer to the semi-pixel point of the image brightness component. Semi-pixel points (such as B, h, m) are obtained by 6 tap filtration of the corresponding whole pixel point, with weight (1/32, -5 / 32, 5/8, 5/8, -5 / 32, 1 / 32). b Calculated by the following formula:
B = ROUND ((E-5F = 20g + 20H-5I + J) / 32) (1)
Similarly, H is obtained by filtered by A, C, G, M, R, and T. Once all pixels of the entire pixel point are calculated from all pixels adjacent (vertical or horizontal direction), the remaining semi-pixel points can be obtained by filtering the half pixel point of 6 vertical or horizontal directions. For example, J is derived from CC, DD, H, M, EE, FF filtering.
After the half pixel point is calculated, on this basis, 1/4 pixel points can be interposed in linearly, as shown in FIG.
Be
Be
Figure 2 Brightness 1/4 pixel inserted
1/4 pixel points (such as A, C, I, K, D, F, N, Q) are obtained from adjacent pixels, such as
A = ROUND ((G + B) / 2) (2)
The remaining 1/4 pixel points (P, R) are linearly interposed from a pair of diagonal half-pixel points, such as e, obtained from B and h. Accordingly, the motion vector of 1/8 pixel precision of the chromaticity component is also interposed through the entire linear interpolation, as shown in FIG.
Be
Be
Figure 3 Chroma 1/8 pixels insert
in:
A = ROUND ([8-DX) (8-DY) A + DX (8-DY) B + (8-DX) DYC + DX DYD] / 64) (3)
FME module algorithm principle and hardware implementation, reading full text, technology area
Hardware simulation commissioning based on embedded systems
MS1830 chip brief description of HDMI turns CVBS function analysis
MS9282 chip introduction function is VGA to HDMI
Detailed integrated transceiver brings SWAP solutions
How to use DFT APP for hardware simulation
Our other product:
