Inter prediction optimization method based on template matching and multi-reference prediction block fusion

Abstract

The application discloses an interframe prediction optimization method based on template matching and multi-reference prediction block fusion, and belongs to the field of video coding, which comprises the following steps: obtaining a current prediction block of a video frame; searching for a prediction block matched with the current prediction block in a current reference frame based on a template matching technology of an encoder; obtaining an additional reference frame, searching for a prediction block matched with the current prediction block in the additional reference frame, and obtaining a candidate prediction block; performing a weighted fusion process on the current prediction block, the prediction block matched with the current prediction block searched in the current frame and the candidate prediction block to generate a new prediction result; comparing and analyzing the original current prediction block result and the new prediction result to select a final prediction result; and adding the final prediction result and residual data based on a decoder to reproduce the generation process of the final prediction result of the encoder to reconstruct a decoded video frame. The application generates prediction blocks by using the same algorithm at the encoding end and the decoding end, reduces the transmission of motion vectors and lowers the code rate overhead.

Description

Technical Field

[0001] This invention belongs to the field of inter-frame prediction optimization technology in the field of video coding, and particularly relates to an inter-frame prediction optimization method based on template matching and multi-reference prediction block fusion. Background Technology

[0002] The goal of video coding is to reduce the storage space and transmission bandwidth requirements of video data through compression techniques while maintaining video quality. Inter-frame prediction is one of the core coding tools in video coding. It improves compression efficiency by analyzing the correlation between adjacent frames in a video sequence and utilizing temporal redundancy. Inter-frame prediction mainly relies on motion estimation and motion compensation techniques to reduce the amount of encoded data by accurately predicting and describing the motion of objects between frames.

[0003] With the increasing diversity and demands of video applications, traditional inter-frame prediction techniques face the dual challenges of prediction accuracy and coding efficiency when dealing with complex scenes. Especially in situations involving rapid motion, occlusion, and scene transitions, existing technologies often struggle to achieve ideal prediction results.

[0004] Currently, multi-reference frames (MFRs) have been introduced into video coding standards. For example, coding standards like H.264 / AVC and HEVC support MFRs, meaning the encoder can use multiple reference frames to predict the current block. However, while MFRs significantly improve prediction accuracy, they also face challenges: the introduction of multiple reference frames requires additional bits to transmit the index and motion vector of each reference frame, leading to an increase in video bitrate. Furthermore, the use of multiple reference frames increases the computational cost of motion estimation, especially in multi-frame scenarios, potentially resulting in higher coding overhead.

[0005] Template matching is a technique for estimating motion vectors by comparing the similarity between the current block and image blocks at different locations in a reference frame. It typically faces the following drawbacks: template matching requires extensive computation at different locations within the reference frame, making it time-consuming. In scenes with complex backgrounds or rapid motion, template matching may fail to effectively capture complex motion information, resulting in lower matching accuracy.

[0006] In recent years, deep learning has made some progress in video coding. For example, motion estimation methods based on deep convolutional neural networks (CNNs) or recurrent neural networks (RNNs) have been proposed. However, deep learning methods also have certain problems: training and inference of deep learning models require a large amount of computational resources, especially in high-resolution video coding, where the computational burden is very heavy. Deep learning models typically require a large amount of training data and cannot adapt to all video scenarios, especially for some special video content, where the model's generalization ability may be limited.

[0007] To address the problems in the existing technologies, there is an urgent need to propose an inter-frame prediction optimization method based on template matching and multi-reference prediction block fusion. Summary of the Invention

[0008] To address the aforementioned technical problems, this invention proposes an inter-frame prediction optimization method based on template matching and multi-reference prediction block fusion, thereby resolving the issues present in the prior art.

[0009] To achieve the above objectives, this invention provides an inter-frame prediction optimization method based on template matching and multi-reference prediction block fusion for video coding, comprising the following steps:

[0010] Get the current prediction block of the video frame;

[0011] The encoder searches for a prediction block that matches the current prediction block in the current reference frame using template matching technology.

[0012] Obtain additional reference frames, and search for prediction blocks that match the current prediction block in the additional reference frames to obtain candidate prediction blocks;

[0013] For the current prediction block, the candidate prediction blocks that match the current prediction block are searched in the current reference frame and then weighted and fused to generate a new prediction result.

[0014] The original current prediction block result and the new prediction result are compared and analyzed, and the final prediction result is selected;

[0015] The encoder's final prediction result generation process is reproduced based on the decoder. Then, the final prediction result is added to the residual data to reconstruct the decoded video frame.

[0016] Optionally, the process of obtaining the current prediction block of a video frame includes:

[0017] The video frame is divided into several coding blocks. Motion estimation is performed on the current coding block based on the encoder to obtain the corresponding motion vector. The current reference frame of the current coding block is obtained. Based on the motion vector, the current prediction block is extracted from the current reference frame through motion compensation.

[0018] Alternatively, the formula for the new prediction result is expressed as follows:

[0019] ,

[0020] Where P0 is the current prediction block, P1 is the prediction block that matches the current prediction block found in the current reference frame, Pi is the prediction block that matches the current prediction block found in the additional reference frame, and Pnew is the new prediction result.

[0021] Optionally, the process of comparing and analyzing the original current prediction block results and the new prediction results to select the final prediction result includes:

[0022] Evaluation metrics for the original current prediction block result and the new prediction result are generated based on the comparison rate distortion cost method. The evaluation metrics are a combination of prediction quality and coding cost. If the evaluation metric of the new prediction result is less than the evaluation metric of the original current prediction block result, the new prediction result is taken as the final prediction result. If the evaluation metric of the new prediction result is greater than or equal to the evaluation metric of the original current prediction block result, the original current prediction block result is taken as the final prediction result.

[0023] Optionally, the process of generating the final prediction result of the encoder based on the decoder includes:

[0024] The decoder receives the reference frame index, residual data, and reference frame data sent by the encoder to each coded block; based on the reference frame index, it uses motion compensation technology to reproduce the current prediction block generated during encoding; and then uses the current prediction block as a template to reproduce the process of the encoder generating the final prediction result.

[0025] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0026] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0027] The present invention also provides a computer program product, including a computer program, characterized in that the computer program, when executed by a processor, implements the steps of the method.

[0028] Compared with the prior art, the present invention has the following advantages and technical effects:

[0029] The method proposed in this invention first obtains the current prediction block of a video frame; then, based on the encoder, it searches for prediction blocks that match the current prediction block in the current reference frame using template matching technology; additional reference frames are obtained, and prediction blocks that match the current prediction block are searched in the additional reference frames to obtain candidate prediction blocks; the current prediction block, the prediction blocks that match the current prediction block found in the current reference frames, and the candidate prediction blocks are subjected to weighted fusion processing to generate a new prediction result; the original current prediction block result and the new prediction result are compared and analyzed to select the final prediction result; the generation process of the encoder's final prediction result is reproduced based on the decoder, and then the final prediction result is added to the residual data to reconstruct the decoded video frame.

[0030] This invention reduces motion vector transmission and lowers bitrate overhead by using the same algorithm to generate prediction blocks at both the encoding and decoding ends. It improves prediction accuracy by fusing prediction blocks from multiple reference frames, performing particularly well in complex motion and background scenes. Furthermore, it employs template matching technology to further optimize motion vectors, enhancing the reconstruction quality of the prediction blocks. Finally, the decoding end of this invention does not require the transmission of motion vectors; the prediction block generation process at the encoding end can be reproduced using only the reference frame index, simplifying the decoding process. Attached Figure Description

[0031] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0032] Figure 1 This is a flowchart of the encoding end of an embodiment of the present invention;

[0033] Figure 2 This is a flowchart of template matching search and multi-reference frame optimization according to an embodiment of the present invention;

[0034] Figure 3 This is a flowchart of the PerformAdvancedInterPrediction method according to an embodiment of the present invention;

[0035] Figure 4 This is a flowchart of the decoding end in an embodiment of the present invention. Detailed Implementation

[0036] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0037] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0038] Example 1

[0039] This embodiment aims to propose a novel inter-frame prediction optimization algorithm. By combining template matching technology and multi-reference frame fusion, it optimizes motion vectors, reduces bitrate overhead, and improves prediction accuracy. Specific objectives are as follows:

[0040] Reduced bitrate overhead: By reproducing the motion vector optimization process at the decoding end, the extra transmission of motion vectors and reference frame indexes is avoided, thereby significantly reducing bitrate overhead.

[0041] Improve prediction accuracy: Motion vectors are optimized through template matching technology, and candidate prediction blocks are generated in multiple reference frames. Multiple prediction blocks are then fused by weighted fusion to improve the reconstruction accuracy of the prediction blocks. This is especially effective in capturing details in complex motion scenes.

[0042] Reduced computational complexity: Despite using template matching and multiple reference frames, the algorithm has moderate computational complexity and can be accelerated by hardware, meeting the needs of efficient video coding.

[0043] Enhanced adaptability to complex scenes: It can significantly improve prediction accuracy in complex backgrounds and fast-moving scenes, ensuring good performance in various video content.

[0044] To achieve the above objectives, this embodiment proposes an inter-frame prediction optimization method based on template matching and multi-reference prediction block fusion, including the following steps:

[0045] Get the current prediction block of the video frame;

[0046] The encoder searches for a prediction block that matches the current prediction block in the current reference frame using template matching technology.

[0047] Obtain additional reference frames, and search for prediction blocks that match the current prediction block in the additional reference frames to obtain candidate prediction blocks;

[0048] For the current prediction block, the candidate prediction blocks that match the current prediction block are searched in the current reference frame and then weighted and fused to generate a new prediction result.

[0049] The original current prediction block result and the new prediction result are compared and analyzed, and the final prediction result is selected;

[0050] The encoder's final prediction result generation process is reproduced based on the decoder. Then, the final prediction result is added to the residual data to reconstruct the decoded video frame.

[0051] This embodiment is feasible and applicable to the inter-frame mode ETM_INTER_ME, and can be applied in both LDP and LDB modes. The main features of this embodiment include:

[0052] (1) Extend to multiple reference frames to generate more prediction blocks through additional reference frames.

[0053] (2) The motion vector of the traditional inter-frame prediction block is further optimized by using template matching technology (TM).

[0054] (3) Weighted fusion of all candidate prediction blocks improves the reconstruction quality of the final prediction block.

[0055] (4) The decoding end does not need to receive motion vectors; it only needs to refer to the frame index to complete the prediction block reconstruction.

[0056] As a specific example, in LDP mode:

[0057] like Figure 1 As shown, the encoding process can be divided into the following main steps:

[0058] This embodiment uses the PerformAdvancedInterPrediction method for initial motion estimation and template matching for multi-reference frame optimization. Figure 2 The example image is for list0 (the same applies to list1 in LDB mode). Figure 3 The flowchart for the PerformAdvancedInterPrediction method includes the following steps:

[0059] The video frame is divided into several coding blocks. Motion estimation is performed on the current coding block (current pu) based on the encoder to obtain a preliminary motion vector (originalMv). The optimal motion vector (originalMv) and the corresponding reference frame (originalRefIdx) are obtained from the current coding block (current pu). The current prediction block P0 is obtained through traditional motion compensation. Using P0 as the reference template and the current motion vector (originalMv) as the starting point, the template matching technique (deriveTMMv) is used to search for the best matching prediction block P1 in the current reference frame (originalRefIdx).

[0060] deriveTMMv

[0061] (pu, / / Current PU unit)

[0062] true, / / Whether to allow interpolation search

[0063] refinedCost, / / The best cost returned

[0064] (RefPicList)refList, / / Current reference list

[0065] originalRefIdx[refList], / / Current reference frame index

[0066] TM_MAX_NUM_OF_ITERATIONS, / / Maximum number of iterations

[0067] refinedMV, / / Returns the best motion vector

[0068] nullptr / / Optional: additional parameters;

[0069] Furthermore, the input is: using the current predicted block generated by the current optimal motion vector as the search target, and searching in the current reference frame (originalRefIdx) for the block that best matches this current predicted block.

[0070] Furthermore, the output is: the motion vector (refinedMV) at the best matching position is used to generate a new prediction block.

[0071] PelUnitBufnewPredBuf= m_tmpStorageLCU.getBuf(UnitAreaRelative(cu,pu));

[0072] pu.mv[REF_PIC_LIST_0] = refinedMV;

[0073] motionCompensation(pu, newPredBuf, REF_PIC_LIST_0);

[0074] Using the new motion vector (refinedMV) obtained from the search, motion compensation is performed in the same reference frame, and the resulting newPredBuf is a new prediction block that is most similar to the current prediction block.

[0075] An feasible multi-reference frame optimization and weighted fusion process includes:

[0076] For the reference list, try using up to 3 additional reference frames. For each additional reference frame, check its validity, optimize the motion vectors, and generate a prediction block.

[0077] Using the same method, P0 is used to find three similar prediction blocks P2, P3, and P4 in frames T-2, T-3, and T-4 respectively. Finally, these candidate prediction blocks are fused to obtain a new prediction result.

[0078] The formula for the new prediction result is expressed as follows:

[0079] ,

[0080] Where P0 is the current prediction block, P1 is the prediction block that matches the current prediction block found in the current reference frame, Pi is the prediction block that matches the current prediction block found in the additional reference frame, and Pnew is the final prediction result.

[0081] The feasible process of comparing and analyzing the original current prediction block results and the new prediction results to select the final prediction result includes:

[0082] Evaluation metrics for the original current prediction block result and the new prediction result are generated based on the comparison rate distortion cost method. The evaluation metrics are a combination of prediction quality and coding cost. If the evaluation metric of the new prediction result is less than the evaluation metric of the original current prediction block result, the new prediction result is taken as the final prediction result. If the evaluation metric of the new prediction result is greater than or equal to the evaluation metric of the original current prediction block result, the original current prediction block result is taken as the final prediction result.

[0083] As an feasible approach, the fused final prediction block Pnew and the current prediction block P0 are compared using the CompareRDCost method to weigh prediction quality (distortion) and coding cost (code rate). The prediction block with the smaller RDCost is then used as the coding result and fed into subsequent coding steps. This prediction block not only improves prediction accuracy but also reduces code rate overhead by reducing the transmission of motion vectors.

[0084] Introduction to the CompareRDCost method:

[0085] The specific implementation is as follows:

[0086] (1) Calculate distortion:

[0087] The function calculates the distortion of the two prediction modes, using the sum of squared differences (SSD) as the distortion metric.

[0088] Calculate the difference between the original image pixel values and the two predicted pixel values, and accumulate the square of the difference into the original signal (originalDist) and the reconstructed signal (newDist).

[0089] (2)Calculate and compare RDCost:

[0090] Use m_pcRdCost to calculate the RD Cost of the two prediction modes.

[0091] originalRDCost = m_pcRdCost->calcRdCost(0, originalDist); Calculate the RD Cost of the original prediction. The bit rate is set to 0 because the original prediction mode is the default mode and no additional bits are required for encoding.

[0092] newRDCost = m_pcRdCost->calcRdCost(1, newDist); Calculate the RD Cost of the new prediction. The bit rate is set to 1 because 1 bit is required to represent the selection of the new prediction mode.

[0093] (3)Return the result:

[0094] return (newRDCost < originalRDCost); If the RD Cost of the new prediction is less than the RD Cost of the original prediction, return true, indicating that the new prediction mode is used; otherwise, return false, indicating that the original prediction mode should be used.

[0095] If true is returned, set newpredictionflag to true; otherwise, set it to false for subsequent writing to the bitstream.

[0096] Write newpredictionflag to the bitstream:

[0097] By performing simple context-free encoding in CABAC (EP = Bypass Mode), directly encode the input value as 1 bit without relying on probability modeling.

[0098] m_BinEncoder.encodeBinEP(cu.newpredictionflag? 1 : 0);

[0099] Write newpredictionflag to the bitstream and wait for the decoding end to read it correspondingly.

[0100] As a specific embodiment, extend it to the LDB mode: ​After initially obtaining the current prediction block P0, it is determined whether it is in LDB mode. If it is LDB, two originalMv values ​​will be obtained from list0 and list1 respectively. Subsequent searches will then be performed to obtain Pnew0 and Pnew1.

[0102] .

[0103] Decoding end:

[0104] The main task of the decoder is, without receiving motion vectors (MV), to reproduce the current prediction block generated during encoding using motion compensation techniques based on the reference frame index, residual data, and reference frame data transmitted from the encoder. Then, using the current prediction block as a template, the decoder reproduces the process of generating the final prediction result from the encoder. The decoder flowchart is as follows: Figure 4 As shown.

[0105] As a specific example, the process of reproducing the prediction block generation at the encoding end at the decoding end typically includes the following steps:

[0106] Receive encoded information:

[0107] Reference Frame Index: The encoder sends the reference frame index for each coded block to inform the decoder which reference frames should be used for prediction.

[0108] Residual: The decoder uses the transmitted residual information to recover the error between the prediction and the actual encoding.

[0109] Reference frame data: The decoding end loads the required reference frame based on the reference frame index.

[0110] Generate initial prediction blocks:

[0111] In Inter mode, the decoder uses motion compensation techniques to reconstruct the current prediction block P0 generated during encoding, based on the reference frame index and block position information (i.e., the block position during encoding). This current prediction block P0 is then obtained as the initial generation result of the prediction block.

[0112] Determine whether to apply the new prediction method based on the value of newpredictionflag:

[0113] If newpredictionflag is true, apply the PerformAdvancedInterPrediction method to generate Pnew; otherwise, use the current prediction block P0.

[0114] PerformAdvancedInterPrediction method:

[0115] Template matching:

[0116] The decoder repeats the template matching process performed by the encoder, but without transmitting motion vectors. The decoder selects a reference frame using the reference frame index and uses the current predicted block P0 as the template for matching within that reference frame.

[0117] Specifically, the decoder uses this template to compare it with different positions in the reference frame, calculates the matching degree (distortion value), and optimizes the motion vector. The template matching method at the decoding end is completely consistent with that at the encoding end.

[0118] Multi-reference frame fusion:

[0119] The decoder uses the current reference frame to generate prediction blocks, and performs prediction using multiple reference frames. Based on the reference frame index transmitted by the encoder, the decoder generates multiple candidate prediction blocks (e.g., P1, P2, ..., Pn) through template matching and motion compensation.

[0120] Then, the decoding end uses the same weighted fusion strategy as the encoding end to fuse the prediction blocks generated from multiple reference frames to obtain the final prediction block.

[0121] The formula for weighted fusion is as follows:

[0122] ,

[0123] In this way, the decoder reproduces the prediction block obtained by the encoder through multi-reference frame optimization.

[0124] In a specific implementation, after the final fused prediction block Pnew is generated, the decoder adds it to the residual data to obtain the final reconstructed block. This reconstructed block is consistent with the prediction block and the actual image content from the encoder. Finally, multiple prediction blocks are reconstructed to form a complete decoded video frame.

[0125] As a specific embodiment, the present invention avoids the transmission of motion vectors: In the prior art, inter-frame prediction typically relies on transmitting motion vectors to help the decoder reconstruct the prediction block, which consumes a large number of bits and increases the bit rate. In contrast, the present invention completely avoids the transmission of motion vectors by using the same prediction block generation algorithm at the decoder as at the encoder, thereby significantly reducing the number of transmitted bits and lowering the bit rate overhead.

[0126] As a specific embodiment, the present invention improves prediction accuracy through multi-reference frame fusion: Traditional video coding methods typically use only a single reference frame for motion estimation, which may fail to fully capture the motion characteristics in complex scenes, resulting in insufficient prediction accuracy. The present invention, however, uses multiple reference frames to generate multiple candidate prediction blocks and improves the accuracy of the final prediction block by weighted fusion of these prediction blocks. Especially in complex motion and high dynamic scenes, it can significantly improve prediction accuracy and reconstruction quality.

[0127] As a specific embodiment, this invention reduces coding complexity: Although some existing multi-reference frame methods can improve prediction accuracy, they typically increase coding complexity, especially in motion estimation and selection of multiple reference frames. This invention optimizes the algorithm by using template matching to select and fuse multiple reference frames, avoiding complex motion vector transmission while maintaining low computational complexity, making the scheme more feasible in hardware implementation.

[0128] As a specific embodiment, the present invention can optimize the performance of the decoding end: since the decoding end does not need to receive motion vectors, but generates prediction blocks through reference frame indexes, the decoding process is simplified. This not only reduces the burden of data transmission, but also ensures the consistency of the encoding and decoding processes, further improving the reconstruction quality of the decoding end, especially in low-latency and real-time video coding applications.

[0129] As a specific embodiment, the present invention can adapt to complex scenarios: it can provide better prediction accuracy in scenes with fast movement and complex backgrounds, and is suitable for application scenarios requiring efficient video compression and low latency, such as high-resolution video compression, low-bitrate video encoding, and real-time video communication. Compared with existing technologies, it can better adapt to the needs of dynamically changing scenes and high-resolution videos.

[0130] Example 2

[0131] This embodiment also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0132] Example 3

[0133] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0134] Example 4

[0135] This embodiment also provides a computer program product, including a computer program, characterized in that the computer program implements the steps of the method when executed by a processor.

[0136] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An inter prediction optimization method based on template matching and multi-reference prediction block merging, characterized in that, The method comprises the following steps: obtaining a current prediction block of a video frame; searching, by an encoder, a prediction block matching the current prediction block in a current reference frame by a template matching technique; obtaining an additional reference frame, searching a prediction block matching the current prediction block in the additional reference frame to obtain a candidate prediction block; performing a weighted fusion process on the current prediction block, the prediction block matching the current prediction block searched in the current reference frame, and the candidate prediction block to generate a new prediction result; comparing and analyzing the original current prediction block result and the new prediction result to select a final prediction result; reproducing, by a decoder, a generation process of the final prediction result of the encoder, and then adding the final prediction result and residual data to reconstruct a decoded video frame.

2. The method of claim 1, wherein the process of obtaining the current prediction block of the video frame comprises: dividing the video frame into a plurality of coding blocks, performing motion estimation on a current coding block by the encoder to obtain a corresponding motion vector, obtaining a current reference frame of the current coding block, and extracting the current prediction block from the current reference frame by motion compensation based on the motion vector.

3. The method of claim 1, wherein the new prediction result is represented by the following formula: wherein P0 is the current prediction block, P1 is the prediction block matching the current prediction block searched in the current reference frame, Pi is the prediction block matching the current prediction block searched in the additional reference frame, and Pnew is the new prediction result.

4. The method of claim 1, wherein the process of comparing and analyzing the original current prediction block result and the new prediction result to select the final prediction result comprises: ， generating evaluation indexes of the original current prediction block result and the new prediction result based on a rate-distortion cost comparison method, the evaluation indexes being a combination of prediction quality and encoding cost; when the evaluation index of the new prediction result is smaller than the evaluation index of the original current prediction block result, the new prediction result is taken as the final prediction result; when the evaluation index of the new prediction result is greater than or equal to the evaluation index of the original current prediction block result, the original current prediction block result is taken as the final prediction result.

5. The method of claim 1, wherein the process of reproducing, by the decoder, the generation process of the final prediction result of the encoder comprises: the decoder receiving reference frame indexes, residual data, and reference frame data sent by the encoder to each coding block; reproducing the current prediction block generated during encoding based on the reference frame indexes by using a motion compensation technique, and then taking the current prediction block as a template to reproduce the process of generating the final prediction result by the encoder. The processor executes the computer program to implement the steps of the method of any one of claims 1-5. The computer program is executed by the processor to implement the steps of the method of any one of claims 1-5. The computer program is executed by the processor to implement the steps of the method of any one of claims 1-5. The computer program is executed by the processor to implement the steps of the method of any one of claims 1-5.

6. A computer apparatus comprising a memory, a processor, and a computer program stored on the memory, wherein the computer program, when executed by the processor, causes the processor to perform the method of any one of claims 1 to 5. ​ 7. A computer-readable storage medium having stored thereon a computer program, characterized in that, ​ 8. A computer program product comprising a computer program, characterized in that, ​

Images