A fast decision-making method based on three-step search of loss function fitting in HEVC intra-frame coding angle mode

Abstract

In response to the limitations of the existing technology, the present invention proposes a fast decision-making method for a three-step search of a loss value function fitting in the angle mode of HEVC intra-frame coding. Based on the assumption that images have structure, the image will not produce huge loss value fluctuations due to small angle deviations during angle prediction, so the image is modeled as a locally near-continuous loss value curve. The fast decision-making method of the three-step search of the loss value function fitting is used to further accelerate the predictive coding. The method can reduce the modes that need to be traversed in the coarse mode decision in intra-frame coding from 35 angle prediction modes to no more than 12, and at least to 8. The coarse mode decision time can be reduced by 53% while keeping the BD-Rate at no more than 0.7% and with little impact on the PSNR.

Description

Technical Field

[0001] The present invention relates to the technical field of rapid decision-making for intra-frame angle mode prediction in the new generation of high-efficiency video coding standard H.265 / HEVC (High Efficiency Video Coding), and more particularly to a rapid decision-making method for a three-step search of loss value function fitting in HEVC intra-frame coding angle mode. Background Art

[0002] With the increasing demand for higher video resolutions and reduced video transmission bandwidth and storage space, the compression performance of the previous-generation video codec standard, H.264 / AVC, is gradually failing to meet this demand. This is especially true for videos with resolutions of 2k and above, which still occupy a considerable amount of data space even after encoding using the H.264 standard. Therefore, the next-generation high-efficiency video codec standard, H.265 / HEVC, was proposed, and HEVC officially became an international standard in January 2013. Compared to H.264, HEVC improves compression efficiency for 1k resolution videos by approximately 50%, resulting in improved video quality while maintaining the same network bandwidth. However, its encoding complexity is also significantly increased, necessitating a reduction in HEVC encoding time to enhance its practicality.

[0003] H.265's compression coding methods include intra-frame compression and inter-frame compression. Inter-frame compression aims to eliminate temporal correlations between frames in a video sequence, while intra-frame prediction aims to eliminate spatial correlations within video frames. By eliminating correlations between coded blocks within a frame, redundancy in the coded image is reduced. Because reference and prediction can only be performed within the frame, the bitrate of intra-frame coded frames is often several to dozens of times higher than that of inter-frame coded frames.

[0004] H.265 increases the number of intra-frame prediction modes from 9 in H.264 to 35. This improves compression prediction accuracy but also increases encoding time. In video coding, the rate-distortion loss function (RDL) is often used as a metric. The rate-distortion optimization process is quite complex, and performing a complete RDL optimization for all 35 angular prediction modes would significantly increase encoding time. To reduce encoding time, H.265 uses a coarse mode decision process to accelerate this process. The residual coefficients predicted from each of the 35 angular prediction modes are subjected to a Hadamard transform and then summed to obtain the SATD. This value, which reflects the bitrate to a certain extent, is then compared with the SATD values ​​to select the modes with the smaller SATD values ​​for the rate-distortion optimization process, thus accelerating intra-frame angular prediction. However, fully iterating through all 35 coarse mode decisions still consumes considerable time, making acceleration of this process essential.

[0005] The Chinese invention patent with an announcement date of March 19, 2019: A method and system for determining an optimal intra-frame prediction mode attempts to obtain a sampling angle mode by sampling the angle mode of the current PU (prediction unit) according to a certain step size strategy (variable step size strategy or uniform step size strategy), avoiding traversal calculations of 35 intra-frame prediction modes, thereby reducing the amount of calculation, shortening the screening time of the angle mode of the current PU, and simplifying the calculation steps; further, the scheme determines the main sampling direction of the current PU based on the texture information of the current PU or the optimal intra-frame prediction mode of the upper or adjacent upper PU. The distortion degree of the sampling main direction and the angle mode nearby is relatively small. Sparse and dense sampling is performed around the sampling main direction according to the variable step size strategy, in order to narrow the sampling range of the angle mode, reduce the complexity of the calculation, and ensure the probability of obtaining the angle mode with the lowest rate-distortion cost. However, since PU division is precisely because different sub-directions may exist in the sub-PU blocks after quadtree division, if the main directions of the sub-PU blocks have large angular differences, using the current upper-layer PU to determine and sample the main direction may result in the direction being unable to converge to a more optimal angular prediction mode. If the adjacent upper-layer PU is used as a reference factor for predicting the main direction of the current PU, the adjacent upper-layer PU may make an erroneous main direction prediction due to the adoption of this scheme, thereby further leading to an erroneous estimation of the main direction of the current PU, and ultimately causing the error to spread to the entire coded image sequence. However, this scheme does not have preventive measures for erroneous estimation, so the existing technology still has certain limitations. Summary of the Invention

[0006] In view of the limitations of the existing technology, the present invention proposes a fast decision-making method for a three-step search of the loss value function fitting in the HEVC intra-frame coding angle mode. The technical solution adopted by the present invention is:

[0007] A fast decision method for three-step search of loss function fitting in HEVC intra-frame coding angle mode, comprising the following steps:

[0008] S1, obtaining current encoding information; according to the current encoding depth, selecting an initial mode list from 33 HEVC intra-frame coding angle modes excluding mode 0 and mode 1 representing DC and Planner modes;

[0009] S2, calculating an estimated rate-distortion loss value list based on the initial pattern list;

[0010] S3, expanding a pattern loop list and a loss value loop list from the estimated rate-distortion loss value list;

[0011] S4, performing a first search: selecting the smallest loss value in the estimated rate-distortion loss value list as the initial minimum value, obtaining the angle pattern corresponding to the initial minimum value according to the pattern loop list and the loss value loop list, obtaining the angle patterns and corresponding loss values ​​adjacent to the initial minimum value, and forming a set of fitting point groups using the three angle patterns and loss values ​​as coordinates.

[0012] S5, performing a second search: using the fitting point group to fit a one-variable quadratic loss value function and its coefficients, using the fitted coefficients to obtain the coordinates of the symmetry axis of the one-variable quadratic loss value function as the angle mode value of the second search step and calculating the corresponding loss value, thereby constructing the lowest point of the one-variable quadratic loss value function;

[0013] S6, performing the third search step: determining the position of the lowest point of the univariate quadratic loss value function, constructing a candidate pattern list based on the determination result, and calculating the loss value of the candidate pattern list;

[0014] S7, compare the loss values ​​of all angle modes searched in steps S4 to S6, select several angle modes for rate-distortion optimization based on the loss values, and compare the angle mode with the lowest rate-distortion loss value in the optimization results as the best mode.

[0015] Compared with the existing technology, the present invention is based on the assumption that images are structural. It uses the fact that images will not produce huge loss value fluctuations due to small angle deviations when performing angle prediction, and thus models them as a locally myopically continuous loss value curve. The fast decision method of three-step search by fitting the loss value function is used to further accelerate the prediction coding. It can reduce the modes that need to be traversed in the coarse mode decision in intra-frame coding from 35 angle prediction modes to no more than 12, and at least to 8. It can reduce the coarse mode decision time by 53% without affecting the PSNR at a BD-Rate of no more than 0.7%.

[0016] As a preferred solution, the initial pattern list ML1 is sampled and obtained in the following manner:

[0017]

[0018] As a preferred solution, in step S3, the following pattern loop list ML1 is expanded from the estimated rate-distortion loss value list CL1: ext And the loss value loop list CL1 ext :

[0019] ML1 ext={ML1[n],ML1[0]...ML1[i],...ML1[n],ML1[0]}, i∈n-1;

[0020] CL1 ext ={CL1[n],CL1[0]...CL1[i],...CL1[n],CL1[0]}, i∈n-1;

[0021] Wherein, n is the number of candidate modes in the initial mode list ML1.

[0022] Furthermore, in step S3, the pattern loop list ML1 is further processed according to the following formula: ext The following compensations were made:

[0023] ML1 ext ={ML1 ext [0]-2,...ML1 ext [i]-2,...ML1 ext [n]-2} i∈n+1;

[0024] ML1 ext [0] = ML1 ext [0]-33;

[0025] ML1 ext [n+1]=ML1 ext [n+1]+33.

[0026] Furthermore, the fitting point group is expressed by the following formula:

[0027] T={P Left ,P center ,P Right};

[0028] in:

[0029] P center =(ML1 ext [mIdx],CL1 ext [mIdx]);

[0030] P Left =(ML1 ext [mIdx-1], CL1 ext [mIdx-1]);

[0031] P Right =(ML1 ext [mIdx+1],CL1 ext [mIdx+1])

[0032] mIdx is the subscript corresponding to the initial minimum value in the estimated rate-distortion loss value list, P Left ,P center ,P Right They represent the left point, center point and right point of the initial minimum value respectively.

[0033] Furthermore, in step S6, if the lowest point P of the one-dimensional quadratic loss function is lx Located in P center Below, put P lx Adjacent pattern points are added to the candidate pattern list ML2:

[0034] If P lx Located in P center On the left side, place P lx Add an adjacent pattern point on the left side of P into ML2, and lx The three adjacent pattern points on the right side of are added to ML2; if P lx Located in P center On the right side, place P lx Add the three adjacent pattern points on the left side of P into ML2, and lx Add one adjacent mode point to the right of ML2.

[0035] Furthermore, in step S6: if the lowest point P of the one-dimensional quadratic loss function is lx Located in P center Below, then with P lx The opposite pattern point is searched and added to the candidate pattern list ML2:

[0036] If P lx Located in P center On the left side, place P center The four adjacent pattern points on the right side of are added to ML2; if P lx Located in P center On the right side, place P center The four adjacent pattern points on the left side are added to ML2.

[0037] The present invention also provides the following:

[0038] A rapid decision-making system for a three-step search of a loss value function fitting in an HEVC intra-frame coding angle mode comprises an initial mode list acquisition module, an estimated rate-distortion loss value list acquisition module, a list expansion module, a first search module, a second search module, a third search module, and an optimization module, which are connected in sequence; the first search module is further connected to the estimated rate-distortion loss value list acquisition module; the optimization module is further connected to the first search module and the second search module; wherein:

[0039] The initial mode list acquisition module is used to obtain current encoding information; according to the current encoding depth, select the initial mode list from 33 HEVC intra-frame coding angle modes excluding mode 0 and mode 1 representing DC and Planner modes;

[0040] The estimated rate-distortion loss value list acquisition module is used to calculate the estimated rate-distortion loss value list according to the initial pattern list;

[0041] The list expansion module is used to expand the estimated rate-distortion loss value list into a pattern cycle list and a loss value cycle list;

[0042] The first search module is configured to perform a first search step: selecting the smallest loss value in the estimated rate-distortion loss value list as the initial minimum value; obtaining the angle pattern corresponding to the initial minimum value, and obtaining the angle patterns and corresponding loss values ​​adjacent to the initial minimum value on the left and right sides according to the pattern loop list and the loss value loop list; and forming a set of fitting point groups using the three angle patterns and loss values ​​as coordinates.

[0043] The second search module performs a second search: using the fitting point group to fit a one-variable quadratic loss value function and its coefficients, using the fitted coefficients to obtain the coordinates of the symmetry axis of the one-variable quadratic loss value function as the angle mode value of the second search step and calculate the corresponding loss value, thereby constructing the lowest point of the one-variable quadratic loss value function;

[0044] The third search module performs a third search: determining the position of the lowest point of the univariate quadratic loss value function, constructing a candidate pattern list based on the determination result, and calculating the loss value of the candidate pattern list;

[0045] The optimization module compares the loss values ​​of all angle modes searched by the first search module, the second search module and the third search module, selects several angle modes for rate-distortion optimization according to the size of the loss values, and compares the angle mode with the lowest rate-distortion loss value in the optimization results as the optimal mode.

[0046] A computer storage medium stores a computer program, which, when executed by a processor, implements the steps of the aforementioned fast decision method for loss value function fitting three-step search in HEVC intra-frame coding angle mode.

[0047] A computer device includes a storage medium, a processor, and a computer program stored in the storage medium and executable by the processor. When the computer program is executed by the processor, the computer program implements the steps of the fast decision method for the three-step search of the loss value function fitting in the aforementioned HEVC intra-frame coding angle mode. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of the stages of a rapid decision-making method for a three-step search of a loss value function fitting in an HEVC intra-frame coding angle mode provided by embodiment 1 of the present invention;

[0049] Figure 2 The cyclic angle mode-loss value curve diagram related to the selection process of step S1 of the three-step search for the loss value function fitting in the HEVC intra-frame coding angle mode provided by embodiment 1 of the present invention;

[0050] Figure 3 This is a schematic diagram of the loss value curve fitting process of the second step search in Example 1 of the present invention;

[0051] Figure 4 Schematic diagram of a fast decision-making system for three-step search of loss value function fitting in HEVC intra-frame coding angle mode provided by embodiment 2 of the present invention. DETAILED DESCRIPTION

[0052] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting this patent;

[0053] It should be clear that the embodiments described are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the embodiments of the present application.

[0054] The terms used in the embodiments of the present application are for the purpose of describing specific embodiments only and are not intended to limit the embodiments of the present application. The singular forms "a," "the," and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0055] When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims. In the description of the present application, it should be understood that the terms "first", "second", "third", etc. are only used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0056] In addition, in the description of this application, unless otherwise specified, "plurality" refers to two or more. "And / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, "A and / or B" can mean: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally indicates that the associated objects are in an "or" relationship. The present invention is further described below with reference to the accompanying drawings and examples.

[0057] In order to solve the limitations of the prior art, this embodiment provides a technical solution, which will be further described below in conjunction with the accompanying drawings and embodiments.

[0058] Example 1

[0059] Please refer to Figure 1 A fast decision method for three-step search of loss function fitting in HEVC intra-frame coding angle mode includes the following steps:

[0060] S1, obtaining current encoding information; according to the current encoding depth, selecting an initial mode list from 33 HEVC intra-frame coding angle modes excluding mode 0 and mode 1 representing DC and Planner modes;

[0061] S2, calculating an estimated rate-distortion loss value list based on the initial pattern list;

[0062] S3, expanding a pattern loop list and a loss value loop list from the estimated rate-distortion loss value list;

[0063] S4, performing a first search: selecting the smallest loss value in the estimated rate-distortion loss value list as the initial minimum value, obtaining the angle pattern corresponding to the initial minimum value according to the pattern loop list and the loss value loop list, obtaining the angle patterns and corresponding loss values ​​adjacent to the initial minimum value, and forming a set of fitting point groups using the three angle patterns and loss values ​​as coordinates.

[0064] S5, performing a second search: using the fitting point group to fit a one-variable quadratic loss value function and its coefficients, using the fitted coefficients to obtain the coordinates of the symmetry axis of the one-variable quadratic loss value function as the angle mode value of the second search step and calculating the corresponding loss value, thereby constructing the lowest point of the one-variable quadratic loss value function;

[0065] S6, performing the third search step: determining the position of the lowest point of the univariate quadratic loss value function, constructing a candidate pattern list based on the determination result, and calculating the loss value of the candidate pattern list;

[0066] S7, compare the loss values ​​of all angle modes searched in steps S4 to S6, select several angle modes for rate-distortion optimization based on the loss values, and compare the angle mode with the lowest rate-distortion loss value in the optimization results as the best mode.

[0067] Compared with the existing technology, the present invention is based on the assumption that images are structural. It uses the fact that images will not produce huge loss value fluctuations due to small angle deviations when performing angle prediction, and thus models them as a locally myopically continuous loss value curve. The fast decision method of three-step search by fitting the loss value function is used to further accelerate the prediction coding. It can reduce the modes that need to be traversed in the coarse mode decision in intra-frame coding from 35 angle prediction modes to no more than 12, and at least to 8. It can reduce the coarse mode decision time by 53% without affecting the PSNR at a BD-Rate of no more than 0.7%.

[0068] Specifically, the present invention is based on the assumption that images have structure, and uses the fact that images will not produce huge loss value fluctuations due to small angle deviations when performing angle prediction, thereby modeling them as a local near-sighted continuous loss value curve. Since angle mode 2 is close to angle mode 34, the two angles are connected end to end and the curve is approximately considered as a cyclic continuous loss value curve. Since it is necessary to search for the lowest loss value of the predicted angle mode in intra-frame coding, the curve is sampled at equal intervals to calculate the corresponding loss value. In order to find the lowest point of the curve without increasing the calculation of too many angle modes, the lowest point and the nearest sampling point to the lowest point, a total of three points, are modeled as on-line points of a local univariate quadratic loss value function, and the three coefficients of the univariate quadratic loss value function are calculated by inverting a third-order matrix. The coordinates of the symmetry axis of the quadratic function are used as the starting point of the second step search. If the search point in the second step is indeed smaller than the lowest point in the first step after calculation, it means that the fitting function is basically correct. The third step search is performed on multiple angle patterns adjacent to the search point in the second step. If the search point in the second step is greater than the lowest point in the first step after calculation, it means that the fitting function has deviated. The third step search point selects multiple patterns in the opposite direction of the lowest point in the first step relative to the search point in the second step for angle prediction calculation.

[0069] More specifically, compared with the Chinese invention patent mentioned in the background technology, the solution proposed in the present invention only estimates the prediction loss of the current PU block, and there will be no error diffusion caused by estimation errors in reference to adjacent or upper and lower layer PU blocks. The estimated minimum value is determined at one time through a fitting function, which saves the excessive encoding time caused by multiple sampling, and compares the estimated minimum value with the remaining values. If an erroneous prediction occurs after comparison, the search direction can be changed in time, preventing large encoding losses caused by errors caused by the estimation method.

[0070] As a preferred embodiment, the initial pattern list ML1 is sampled and obtained in the following manner:

[0071]

[0072] Specifically, in step S1, mode 0 and mode 1 representing DC and Planner modes are removed, and equidistant modes are selected from the remaining 33 angle modes, such as Figure 2 The displayed cyclic angle pattern-loss value curve; the selection principle is that the deeper the depth, the more likely it is that the image complexity is higher, and a shorter spacing is used for pattern sampling.

[0073] As a preferred embodiment, in step S3, the following pattern loop list ML1 is expanded from the estimated rate-distortion loss value list CL1: ext And the loss value loop list CL1 ext :

[0074] ML1 ext ={ML1[n],ML1[0]…ML1[i],…ML1[n],ML1[0]}, i∈n-1;

[0075] CL1 ext ={CL1[n],CL1[0]…CL1[i],…CL1[n],CL1[0]}, i∈n-1;

[0076] Wherein, n is the number of candidate modes in the initial mode list ML1.

[0077] In order to consider the range of angle mode values ​​to be 0 to 32, the starting angle mode value 2 can be subtracted from all mode numbers, and the previous and next mode values ​​can be compensated. Therefore, further, in the step S3, the mode loop list ML1 is also calculated according to the following formula: ext The following compensations were made:

[0078] ML1 ext ={ML1 ext [0]-2,…ML1 ext [i]-2,…ML1 ext [n]-2} i∈n+1;

[0079] ML1 ext [0] = ML1 ext [0]-33;

[0080] ML1 ext [n+1]=ML1 ext [n+1]+33.

[0081] Furthermore, the fitting point group is expressed by the following formula:

[0082] T={P Left ,P center ,P Right};

[0083] in:

[0084] P center =(ML1 ext [mIdx],CL1 ext [mIdx]);

[0085] P Left =(ML1 ext [mIdx-1], CL1 ext [mIdx-1]);

[0086] P Right =(ML1 ext [mIdx+1],CL1 ext [mIdx+1])

[0087] mIdx is the subscript corresponding to the initial minimum value in the estimated rate-distortion loss value list, P Left ,P center ,P Right They represent the left point, center point and right point of the initial minimum value respectively.

[0088] Specifically, P center That is, the coordinates composed of the initial minimum value and the corresponding angle pattern; P Left That is, the loss value loop list CL1 is composed of the initial lowest value ext The adjacent loss value on the left and the corresponding angle mode; P Right That is, the loss value loop list CL1 is composed of the initial lowest value ext The adjacent loss values ​​on the right and the corresponding angle patterns.

[0089] In the second step of the search, the set of fitting points is used to fit the quadratic loss value function and its coefficients, and the symmetry axis coordinates of the function are obtained using its coefficients as the coordinates of the lowest point. The coordinates are rounded as the search point mode value of the second step and the loss value is calculated. The fitting process is as follows: Figure 3 As shown (the original image is in color), it is a part of the loss value curve that includes the initial lowest point. The solid line is the true loss value, the dotted line is the fitted quadratic loss value function, the green points (the first, fifth, and tenth points from the left on the solid line) are the initial calculation mode points sampled at equal intervals, and the yellow point (the second point from the left passed by the dotted line) is the lowest point of the fitting curve. The lowest point of the fitting curve will be rounded as the starting point for the second search.

[0090] In the third search step, if the loss of the second search point is less than or equal to the lowest loss value of the initial search point, it means that the local loss value quadratic function fitting is successful, and the adjacent pattern points of the second search point are added to the third search pattern list. If the loss of the second search point is greater than the lowest loss value of the initial search point, it means that the local loss value quadratic function fitting has failed, and the adjacent pattern points of the lowest loss value point of the initial search point in the opposite direction of the second search point are added to the third search pattern list.

[0091] Furthermore, in step S6, if the lowest point P of the one-dimensional quadratic loss function is lx Located in P center Below, put P lx Adjacent pattern points are added to the candidate pattern list ML2:

[0092] If P lx Located in P center On the left side, place P lx Add an adjacent pattern point on the left side of P into ML2, and lx The three adjacent pattern points on the right side of are added to ML2; if P lx Located in P center On the right side, place P lx Add the three adjacent pattern points on the left side of P into ML2, and lx Add one adjacent mode point to the right of ML2.

[0093] Furthermore, in step S6: if the lowest point P of the one-dimensional quadratic loss function is lx Located in P center Below, then with P lx The opposite pattern point is searched and added to the candidate pattern list ML2:

[0094] If P lx Located in P center On the left side, place P center The four adjacent pattern points on the right side of are added to ML2; if P lx Located in P center On the right side, place P center The four adjacent pattern points on the left side are added to ML2.

[0095] Example 2

[0096] A fast decision system for fitting a three-step search of the loss function in HEVC intra-frame coding angle mode, see Figure 4, comprising an initial pattern list acquisition module 1, an estimated rate-distortion loss value list acquisition module 2, a list expansion module 3, a first search module 4, a second search module 5, a third search module 6, and an optimization module 7, which are connected in sequence; the first search module 4 is also connected to the estimated rate-distortion loss value list acquisition module 2; the optimization module 7 is also connected to the first search module 4 and the second search module 5; wherein:

[0097] The initial mode list acquisition module 1 is used to obtain current encoding information; according to the current encoding depth, the initial mode list is selected from 33 HEVC intra-frame coding angle modes excluding mode 0 and mode 1 representing DC and Planner modes;

[0098] The estimated rate-distortion loss value list acquisition module 2 is used to calculate the estimated rate-distortion loss value list according to the initial pattern list;

[0099] The list expansion module 3 is used to expand the estimated rate-distortion loss value list into a pattern cycle list and a loss value cycle list;

[0100] The first search module 4 is configured to perform a first search step: selecting the smallest loss value in the estimated rate-distortion loss value list as the initial minimum value; obtaining the angle pattern corresponding to the initial minimum value, and obtaining the angle patterns and corresponding loss values ​​adjacent to the initial minimum value on the left and right sides according to the pattern loop list and the loss value loop list; and forming a set of fitting point groups using the three angle patterns and loss values ​​as coordinates.

[0101] The second search module 5 performs a second search: using the fitting point group to fit a one-variable quadratic loss value function and its coefficients, using the fitted coefficients to obtain the coordinates of the symmetry axis of the one-variable quadratic loss value function as the angle mode value of the second search step and calculate the corresponding loss value, thereby constructing the lowest point of the one-variable quadratic loss value function;

[0102] The third search module 6 performs a third search: determining the position of the lowest point of the univariate quadratic loss value function, constructing a candidate pattern list based on the determination result, and calculating the loss value of the candidate pattern list;

[0103] The optimization module 7 compares the loss values ​​of all angle modes searched by the first search module 4, the second search module 5 and the third search module 6, selects several angle modes for rate-distortion optimization according to the size of the loss values, and compares the angle mode with the lowest rate-distortion loss value in the optimization results as the optimal mode.

[0104] Example 3

[0105] A computer storage medium stores a computer program, which, when executed by a processor, implements the steps of the fast decision method for three-step search of loss value function fitting in HEVC intra-frame coding angle mode in embodiment 1.

[0106] Example 4

[0107] A computer device includes a storage medium, a processor, and a computer program stored in the storage medium and executable by the processor. When the computer program is executed by the processor, the computer program implements the steps of the fast decision method for the three-step search of the loss value function fitting in the HEVC intra-frame coding angle mode in Example 1.

[0108] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A fast decision method for three-step search of loss function fitting in HEVC intra-frame coding angle mode, characterized in that: The following steps are involved: S1, obtain the current coding information; according to the current coding depth, select the initial mode list from the 33 HEVC intra-frame coding angle modes excluding mode 0 and mode 1 representing DC and Planner modes ; S2, according to the initial pattern list Calculate the estimated rate-distortion loss value list ; S3, based on the estimated rate-distortion loss value list Expand the mode loop list And the loss value loop list ; S4, perform the first search: select the estimated rate distortion loss value list The minimum loss value is used as the initial minimum value, and the list is looped according to the pattern And the loss value loop list , obtain the angle pattern corresponding to the initial minimum value, obtain the angle patterns adjacent to the initial minimum value on the left and right and the corresponding loss values; respectively use the above three angle patterns and loss values ​​as coordinates to form a set of fitting point groups; S5, performing a second search: using the fitting point group to fit a one-variable quadratic loss value function and its coefficients, using the fitted coefficients to obtain the coordinates of the symmetry axis of the one-variable quadratic loss value function as the angle mode value of the second search step and calculating the corresponding loss value, thereby constructing the lowest point of the one-variable quadratic loss value function; S6, perform the third step of searching: determine the position of the lowest point of the univariate quadratic loss function, and construct a candidate pattern list based on the determination result. , for the candidate pattern list Calculate the loss value; S7, comparing the loss values ​​of all angle modes searched in steps S4 to S6, selecting several angle modes for rate-distortion optimization based on the loss values, and comparing the angle mode with the lowest rate-distortion loss value in the optimization results as the best mode.

2. The fast decision-making method for three-step search of loss value function fitting in HEVC intra-frame coding angle mode according to claim 1, characterized in that The initial pattern list Sampling is performed in the following manner: 。 3. The fast decision-making method for three-step search of loss value function fitting in HEVC intra-frame coding angle mode according to claim 1, characterized in that In step S3, the estimated rate-distortion loss value list Expand the following mode loop list And the loss value loop list : ； ； Wherein, n is the initial pattern list The number of candidate patterns in .

4. The fast decision method for three-step search of loss value function fitting in HEVC intra-frame coding angle mode according to claim 3, characterized in that In step S3, the pattern loop list is further processed according to the following formula: The following compensations were made: ； ； 。 5. The fast decision method for three-step search of loss value function fitting in HEVC intra-frame coding angle mode according to claim 3, characterized in that The fitting point group is expressed by the following formula: ； in: ； ； is the subscript corresponding to the initial minimum value in the estimated rate-distortion loss value list, They represent the left point, center point and right point of the initial minimum value respectively.

6. The fast decision method for three-step search of loss value function fitting in HEVC intra-frame coding angle mode according to claim 5, characterized in that In step S6, if the lowest point of the one-dimensional quadratic loss function is lie in Below, Adjacent pattern points are added to the candidate pattern list middle: like lie in On the left side, Add an adjacent pattern point to the left of ,Will Add the three adjacent pattern points on the right side of ;like lie in On the right side, Add the 3 adjacent pattern points on the left side of ,Will Add an adjacent mode point to the right of .

7. The fast decision method for three-step search of loss value function fitting in HEVC intra-frame coding angle mode according to claim 5, characterized in that In step S6: if the lowest point of the one-dimensional quadratic loss function is lie in Below, Search for pattern points in the opposite direction and add them to the candidate pattern list middle: like lie in On the left side, Add the 4 adjacent pattern points on the right side of ;like lie in On the right side, Add the 4 adjacent pattern points on the left side of .

8. A fast decision system for three-step search of loss function fitting in HEVC intra-frame coding angle mode, characterized in that: The system comprises an initial pattern list acquisition module (1), an estimated rate-distortion loss value list acquisition module (2), a list expansion module (3), a first search module (4), a second search module (5), a third search module (6), and an optimization module (7) connected in sequence; the first search module (4) is further connected to the estimated rate-distortion loss value list acquisition module (2); the optimization module (7) is further connected to the first search module (4) and the second search module (5); wherein: The initial mode list acquisition module (1) is used to obtain the current coding information; according to the current coding depth, the initial mode list is selected from the 33 HEVC intra-frame coding angle modes excluding mode 0 and mode 1 representing DC and Planner modes. ; The estimated rate distortion loss value list acquisition module (2) is used to obtain the estimated rate distortion loss value list according to the initial mode list. Calculate the estimated rate-distortion loss value list ; The list expansion module (3) is used to list the estimated rate distortion loss value Expand the mode loop list And the loss value loop list ; The first search module (4) is used to perform the first search: select the estimated rate distortion loss value list The minimum loss value is used as the initial minimum value, and the list is looped according to the pattern And the loss value loop list , obtain the angle pattern corresponding to the initial minimum value, obtain the angle patterns adjacent to the initial minimum value on the left and right and the corresponding loss values; respectively use the above three angle patterns and loss values ​​as coordinates to form a set of fitting point groups; The second search module (5) performs a second search: using the fitting point group to fit a one-dimensional quadratic loss value function and its coefficients, using the fitted coefficients to obtain the coordinates of the symmetry axis of the one-dimensional quadratic loss value function as the angle mode value of the second search step and calculate the corresponding loss value, thereby constructing the lowest point of the one-dimensional quadratic loss value function; The third search module (6) performs the third search step: judging the position of the lowest point of the one-dimensional quadratic loss value function, and constructing a candidate pattern list based on the judgment result , for the candidate pattern list Calculate the loss value; The optimization module (7) compares the loss values ​​of all angle modes searched by the first search module (4), the second search module (5) and the third search module (6), selects several angle modes for rate-distortion optimization according to the size of the loss values, and compares the angle mode with the lowest rate-distortion loss value in the optimization results as the best mode.

9. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the fast decision method for three-step search of loss value function fitting in HEVC intra-frame coding angle mode are implemented.

10. A computer device, characterized in that: The invention comprises a storage medium, a processor and a computer program stored in the storage medium and executable by the processor, wherein when the computer program is executed by the processor, the method for quickly deciding the loss value function fitting three-step search in the HEVC intra-frame coding angle mode as described in any one of claims 1 to 7 is implemented.

Images