Key words:
Estimation algorithm, fast exercise
H. 264 is one of the international standards of the existing most important data compression coding. At the same time, the rapid motion estimation calculation has always been a research hotspot in video compression. In this paper, some fast estimation calculations are premature search directions, which is easy to fall into local minimum points, and lose search accuracy. On the original basis, the method of combining the diamond cross search algorithm (DCS) and threshold estimation is improved. Performance. The experimental results show that the algorithm can be obtained with a smaller search price, and the effect is equivalent to full search (FS) and is superior to the diamond method DS in terms of search speed.
0 Preface
H. 264 The motion estimation algorithm for reducing the redundancy of each frame of video images. Motion Estimation Calculation Law Traditional Search (FS), Three-Step Search (TSS), New Three Search (NTSS), Four Search (FSS), etc .; commonly used diamond search (DS) and asymmetric cross multi-level Six-side grid point motion search algorithm (umhexagons). Motion estimation is the largest module in the entire video encoding, which can account for more than 70% of the entire software encoder operation. Therefore, the complex part of the encoder in the video system depends on the complexity of the motion estimation calculation method.
1 motion estimation calculation method analysis
1.1 Diamond Search (DS) Algorithm
The DS algorithm is a diamond search algorithm, which is used in the MPEG-4 International Standard and the Verification Model, which is the rapid motion estimation calculation method recommended by MPEG-4. The DS algorithm uses two search templates, which are 9 detective large templates and a small template with 5 detection points. When searching, use large template LDSP to match the search area center and around 8 points. When the minimum block error occurs at the center point, the large template LDSP is changed to SDSP, and the matching calculation is performed. At this time, 5 points The MINIMUM Block Distort is an optimal matching point; otherwise, change the center position, still use LDSP repeat.
1.2 Rhombus Cross Search (DCS) Algorithm
The rhombus cross search algorithm (DCS) algorithm is based on the DS algorithm, and the DCS matching template is based on two different search templates, namely, Daxi Cross-shaped (LDCSP) and small rhombus cross-shaped cross-shaped transmissions (SDCSP) ),As shown in Figure 1. Among them, the cross-type can correspond to the vector distribution of the actual motion, and cross-type is to speed up search results.
DCS algorithm has three possible situations when performing motion estimation matching operations:
(1) If the MBD point is located at the center of the LDCSP, the image is stationary, and the DCS algorithm ends in one step;
(2) If the MBD point is located in the LDCSP small cross position, the movement of the image is smaller, and the cross search is repeated in accordance with the SDCSP template based on this.
(3) If the MBD point is located in the LDCSP large cross position, the motion of the image is large, and the cross search is repeated in accordance with the LDCSP template based on this.
It can be seen that the advantage of the DCS algorithm is that it can select the next corresponding search template according to the motion type of the image (such as the above), white adaptation to select the next corresponding search template, so that the search is related to the image content (based on the content of the content), thereby Get good search effects; DCS algorithm search does not have to experience the template by large to small inevitable process, sometimes step by step; search with DCS, the cross-shaped template corresponds to the actual motion vector distribution, cross-shaped The template has accurate "focus" characteristics, which essentially reflects the effective combination of DCS is coarse positioning and accurate positioning.
Be
2 Threshold determination of motion estimation
Most of the video image of motion vector is zero vector or small vector. The motion vector is zero block is called still block; the motion vector is small (in the center of the search window, the two pixels is circular) is called quasi-stop block: Other Sports Blocks. If there is more than 80% of motion vectors, a small block can be seen as still or quenching. Therefore, one threshold T can be provided. When the value of the motion vector is less than T, the SDCSP search method can be used directly, and the optimal point can be found; when the value of the motion vector is greater than T, the optimal point can be found with the LDCSP search algorithm.
The prediction method of the motion vector spatial domain has a motion vector median prediction, the upper block mode motion vector of the spatial domain; the prediction method of the time domain has a predictive reference frame motion vector prediction of the time domain. This paper uses a motion vector median prediction method. According to the left A block adjacent to the current E block, the motion vector of the upper side B and the upper side C block is taken as the prediction motion vector of the current block. as shown in picture 2.
Set the motion vector of the current to encode block E to MVP. If the value MVP ≥ T, the correlation between the various blocks adjacent to the E block is high, indicating that the variation of the region is relatively gentle; if the value of the motion vector is MVP
Our other product:
