Computed tomography (CT) key frame selection method based on complex phase constellation coding

Abstract

The invention provides a CT (Computed Tomography) key frame selection method based on complex phase constellation coding. The CT key frame selection method comprises the following steps: step 1, slice complex phase symbolization: slices are changed into complex phase symbols capable of being rechecked; step 2, performing W-CCPS segmentation on an amplitude-coupled circular phase change point codebook, and changing a selected frame into proved continuous coverage; step 3, irreplaceable pilot frequency selection and transition zone closed-loop frame supplementation are carried out; and step 4, generating CPDE through variable-point driving evidence, and outputting an auditing evidence packet. The method has the advantages that coverage can be verified and closed-loop, cross-protocol robustness and interpretability are achieved, excessive segmentation driven by noise is avoided, the risk of missed reading of a transition belt is reduced, and expandability is achieved.

Description

Technical Field

[0001] This invention belongs to the field of computed tomography volume data processing technology, and particularly relates to a CT keyframe selection method based on complex phase constellation coding. Background Technology

[0002] Computed tomography (CT) scan data often contains hundreds to thousands of slices. Clinical image interpretation and downstream AI tasks (segmentation, detection, report generation, follow-up comparison, etc.) often require automatically selecting sufficiently representative and verifiable keyframes from massive amounts of slices to improve efficiency and reduce the risk of missed readings.

[0003] Existing keyframe selection techniques generally include: 1) Top-K selection based on fixed intervals or underlying statistics (variance, entropy, gradient energy, etc.): The advantage is that it is simple and fast, but it cannot characterize the stage-by-stage structural state of the anatomy as it evolves continuously with the Z-axis, and it is easy to miss selection at structural transitions or over-select in homogeneous regions. 2) Frame-by-frame ranking based on deep feature scoring / attention: Although deep features are introduced, most of them are still scored independently for each frame, lacking overall constraints on the compressible representation and reconstructible / decodeable coverage of the entire sequence; 3) Clustering / Reconstruction-type Representative Sampling: Emphasizes distribution coverage, but often uses the average frame as representative, without explicitly constraining the distinguishability and cross-protocol stability of diagnostic semantics.

[0004] The main drawbacks of existing technologies include: (1) Lack of full sequence coverage provability: Most methods only guarantee that the selected frame looks more informational, but cannot answer whether the remaining unselected slices can be reliably represented / interpreted by these key frames, lacking provable constraints on coverage quality; (2) Sensitive to protocol differences and noise: Different scanning thicknesses, reconstruction kernels, and contrast agent phases can lead to fluctuations in pixel statistics and local features, resulting in unstable frame-by-frame scoring and poor repeatability of keyframe sets; (3) Interpretation remains at a low level of vision: the common interpretation is the gradient / texture contribution map, which cannot form a structured interpretation of which anatomical stages this keyframe represents, which segments it covers, and which stage it is most easily confused with; (4) Redundancy control lacks sequence-level structure: similarity thresholds or distance penalties are mostly empirical, making it difficult to simultaneously balance excessive density (low redundancy) and excessive sparseness (sufficient coverage).

[0005] Therefore, a new keyframe selection mechanism is needed: it forms a verifiable overlay structure at the sequence level and naturally provides evidence consistent with clinical image interpretation logic. Summary of the Invention

[0006] Purpose of the invention: The technical problem to be solved by the present invention is to provide a CT keyframe selection method based on complex phase constellation coding, which addresses the shortcomings of the existing technology.

[0007] The technical problems to be solved by this invention include: (1) How to convert each slice of CT body data into a measurable, comparable and stable anatomical state symbol that is stable to protocol perturbations, so as to organize and cover the slices at the whole sequence level; (2) How to learn an adaptive constellation codebook on the complex plane so that any slice can be mapped to a codebook symbol with low confusion probability, thus achieving decodeable coverage; (3) How to select the minimum number of pilot keyframes from the codebook so that they represent each anatomical stage and have the maximum robustness to noise / reconstruction differences; (4) How to generate structured evidence: including the sequence interval represented by the key frame, decision boundaries, easily confused adjacent stages, and the quantization result of coverage degradation caused by removing the frame.

[0008] This method includes the following steps: Step 1: Slice complex phase symbolization: Transform the slice into a complex phase symbol that can be reconstructed; Step 2: Amplitude-coupled circular change-point codebook segmentation W-CCPS (Weighted-Circular Change-Point Segmentation) transforms the selected frame into a verifiable continuous coverage area; Step 3: Irreplaceable pilot selection and transition band closed-loop frame interpolation; Step 4: Generate CPDE (Change-Point Driven Evidence) and output an auditable evidence package; Step 1 includes: Step 1.1, Metadata Constraint Perturbation Generator (MPG): Transforms the enhancements into a protocol-physically consistent family of perturbations; Read the DICOM (Digital Imaging and Communications in Medicine) metadata M and construct a perturbation family that is physically consistent with the case protocol. ,include: Layer thickness: only in Resampling is performed within a reasonable range; where T represents time. Represents a time variable; Application: The embeddings produced by thin or thick layers should be roughly consistent; otherwise, the slice will be sensitive to layer thickness and have low stability.

[0009] Reconstructing the kernel: Simulate kernel differences using light sharpening and smoothing that satisfy the monotonic constraints of the MTF (Modulation Transfer Function); Application: If the embedding of the same slice is consistent in soft tissue nuclei and slightly sharp nuclei, it indicates that its criticality does not depend on the high-frequency texture of a particular nucleus. Noise: Estimate the equivalent noise based on dose or noise index and inject accordingly; Application: Slices whose vector values ​​change with a change in noise are considered unstable, and their weight in subsequent segmentation and frame selection is reduced; Window width and level: limited to a clinically reasonable range with slight vibration; Application: Stable slices should maintain consistent vector values ​​under window changes; otherwise, their weight in subsequent segmentation and frame selection should be reduced. Step 1.2, Stability (STA): Verifiable reliability is defined using the consistency definition under the perturbation family; Encode the perturbation slice to obtain the feature vector of the same slice under slight changes compatible with the current protocol. Fingerprint consistency among two or more perturbations on the same slice can be obtained using projection sign consistency. , The fingerprint consistency ratio between perturbation pairs, of which Indicates direct proportion; Step 1.3, Event-Based EVB: Use dual anchors to predict rupture phase transitions or structural events; Search within the windows on both sides of the z-axis for the prediction vector that best explains the current slice. To obtain the residual : ; in It is the original slice The embedding vector; Step 1.4, Amplitude Gating (AMG): Combines stable and reliable data with event-critical data to form an amplitude. ; ; Step 1.5, Phase (PHA): Representing anatomical progress using cumulative curvature: , in It is the phase of the anatomical process, along the embedded sequence. The cumulative turning angle of the trajectory represents the scanning progress; in It is the angle between adjacent tangent vectors; mod is the modulo function; Step 1.6, obtain the complex symbols for each layer: , in, It is the complex phase symbolization of the slice, where e represents the natural constant and j represents the imaginary number.

[0010] Step 2 includes: Step 2.1, obtain the amplitude from Step 1. Converted into weights ; Where β is the emphasis coefficient of the weight, controlling the weighting of high-reliability slices (high... The intensity of preference; Step 2.2, weighted circumferential dispersion; Step 2.3, Stability Adaptive Penalty Λ (a piecewise penalty term used to suppress excessive segmentation in low-stability (noise or spurious change) segments): using the intra-segment average stability... (through the first) each interval The average value of STA is obtained, where , They represent the first time. In each segment, slices a and b) adaptively adjust the penalty; Step 2.4: Output the segmented coverage interval.

[0011] Step 2.2 includes: using circumferential statistics to describe whether the phases are concentrated within a certain segment: , in, It is an interval The amplitude-weighted circular dispersion is used to measure whether intervals belong to the same stage. b and b are two different slices; This means placing the phase on the unit circle.

[0012] In step 2.4, N consecutive intervals are obtained. , Indicates the first The coverage areas of slices a and b in each segment.

[0013] Step 3 includes: Step 3.1, Irreplaceability (IRP): Defining Irreplaceability (indicated in the first) Within a segment, if candidate frames are... Removal will cause a decrease in intra-segment consistency evidence (how much), and the largest one (in the first segment) is selected. The one with the greatest irreplaceability within the segment is the dominant frequency. ; Step 3.2, Transitional Confusion Bandwidth (CBW): Calculate the minimum window width near each segment boundary. This ensures that both the narrow and wide transition sides reach a stable threshold. At the border Nearby, expand the window width to both sides. Left window right window Calculate whether the phase concentration on the left and right sides exceeds a threshold (using a fixed threshold method with values ​​of 6 to 8 based on testing). The window is considered stable if and only if the weighted circular phase concentration of the defined window is greater than or equal to the threshold. The corresponding minimum window size is... ; Step 3.3, closed-loop frame interpolation: when Large enough to exceed the threshold Automatically select another event pilot within the transition band. (Add a crucial event frame within the transition band to cover the blurred transition.) It is the risk threshold for triggering closed-loop frame interpolation. If the transition band is deemed too wide (high risk), encrypted sampling is required; if The transition is clear and no frame interpolation is needed.

[0014] In step 3.3, the final set of keyframes is obtained. : .

[0015] In step 4, the following four types of evidence are output: Coverage evidence, including: each interval Dominant frequency intra-segment stability ; Consistency evidence includes: intra-segment amplitude-weighted consistency evidence for each segment; Evidence of necessity, including: evidence of necessity for each pilot segment; Transition risks and closed-loop records, including: each boundary Whether it was triggered, the reason for the trigger, and the event pilot score.

[0016] The present invention also provides an electronic device, including a processor and a memory, the memory storing program code that, when executed by the processor, causes the processor to perform the steps of the method.

[0017] The present invention also provides a storage medium storing a computer program or instructions that, when the computer program or instructions are run on a computer, execute the steps of the method described.

[0018] This invention proposes a novel modeling paradigm: treating CT slice sequences as anatomical state signals evolving along the Z-axis, first mapping each slice to a complex phase symbol on the complex plane, then learning a complex phase constellation codebook, and finally outputting keyframes (pilot frames) and interpretable evidence through pilot-driven decoding coverage. The core of this paradigm is not to weight existing scoring functions, but rather to use full-sequence decodeable coverage as the primary selection principle.

[0019] The present invention has the following beneficial effects: (1) Coverage is verifiable and closed-loop: Through segmented intervals, weighted evidence, and transition zone risk-triggered frame interpolation mechanism, the entire process is traceable and automatically corrected; (2) Robust and interpretable across protocols: Metadata constraints are used to generate perturbations and assess stability, ensuring consistent features across different scanning protocols and explaining why supplementary frames are needed; (3) Avoid noise-driven oversegmentation: Introduce an adaptive penalty term based on the average stability within the segment to effectively suppress incorrect segmentation caused by noise or pseudo-changes; (4) Reduced risk of missed reading in transition zone: When the transition zone is detected to be blurry or too wide, event pilot frames are automatically added at key locations to improve the reliability of boundary recognition; (5) High scalability: The weighting emphasis coefficient, risk threshold and perturbation parameters can be adaptively adjusted according to different examination sites, scanning protocols and clinical indications, and have good generalization ability. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention.

[0021] Figure 2 This is a diagram comparing phase and bandwidth (conventional and ultrawide).

[0022] Figure 3 This is a diagram illustrating the dense coverage intervals of navigation and bandwidth-triggered events. Detailed Implementation

[0023] like Figure 1 As shown, this embodiment of the invention provides a CT keyframe selection method based on complex phase constellation coding, including the following steps: Step 1: Slice complex phase symbolization: Transform the slice into a complex phase symbol that can be reconstructed; The core pain point to be addressed is that traditional keyframe methods suffer from frame drift under different scanning protocols (slice thickness, reconstruction kernel, noise level), making it difficult for doctors to reliably verify or follow up on comparisons. On the other hand, simply performing deep feature or gradient entropy ranking is not accurate enough to determine which segment of the scanning process a particular frame represents.

[0024] The first step of this invention is not scoring, but creating a new intermediate language: complex phase notation; Amplitude: Whether this slice remains stable and reliable under perturbations consistent with the current protocol, and whether it is at a critical point in a structural event / phase transition.

[0025] Phase: This refers to the anatomical progress angle of this slice throughout the entire scan, used to describe the progress and stage boundaries, rather than pixel gradients.

[0026] Step 1.1, Metadata Constraint Perturbation Generation: Transform the enhancements into a protocol-physically consistent family of perturbations; Unlike conventional data augmentation, this invention does not use arbitrary random augmentation. Instead, it reads the DICOM metadata M and constructs a perturbation family that is physically consistent with the protocol of the case. .

[0027] Layer thickness: only in Resampling within a reasonable range; Reconstructing the kernel: Simulate kernel differences using light sharpening / light smoothing that satisfies the MTF monotonic constraint; Noise: Equivalent noise is estimated and injected based on dose / noise index; Window width and window level: limited to a clinically reasonable range such as abdominal window / lung window with slight shaking.

[0028] Step 1.2, Stability (STA): Verifiable reliability is defined using the consistency definition under the perturbation family; Encoding the perturbation slice yields Obtained by projection symbol consistency .

[0029] The fingerprint consistency ratio between perturbation pairs, of which It indicates that they are directly proportional.

[0030] The more stable a slice is under the protocol-consistent perturbation family, the higher its stability STA, indicating that it is more suitable as a representative slice for that stage.

[0031] Step 1.3, Event-based EVB: Use dual anchors to predict rupture and identify stage transition / structural events; Find the prediction vector that best explains the current slice within the windows on both sides of the z-axis. Then look at the residuals Core expression: , If the current layer can be smoothly interpreted by the context on both sides, it is not an event; if it cannot be interpreted (large residuals), it is often at the point of change in the anatomical stage or a structural abrupt change.

[0032] Step 1.4, Amplitude Gating (AMG): Combines stable reliability with event-critical amplitude. ; The magnitudes are not simply added together; rather, they must simultaneously satisfy the requirements of verifiable stability and event-driven phases. STA ensures cross-protocol review; EVB captures stage-specific mutations; The two gating forms amplitude : , By binding robustness and event-driven behavior to the same scalar amplitude, an irreplaceable coupling signal is provided for subsequent segmentation and frame selection.

[0033] Step 1.5, Phase PHA: Representing the anatomical progress using curvature accumulation; The phase is neither the PCA projection angle nor an arbitrary atan2; this invention uses the corner accumulation of the embedded curve to ensure that the phase only reflects the progress of the process and is not affected by intensity scaling: , in It is the angle of rotation between adjacent tangent vectors.

[0034] Step 1.6 yields the complex symbol; Finally, we obtain the complex symbols for each layer: ; This complex symbol simultaneously carries whether the layer is reliable (amplitude) and what position the layer is in in terms of the anatomical process (phase), providing an auditable sequence representation for subsequent segmentation and frame selection that is not found in traditional keyframes.

[0035] Step 2: Amplitude-coupled circular phase-change codebook segmentation (W-CCPS) – turning frame selection into a provable continuous coverage area; The core pain point to be addressed: Many methods only provide K frames, but cannot answer: Who represents the hundreds of slices in the middle? Are there any missed scans in the transition zone? Are the slices fragmented due to noise? The second step of this invention does not involve clustering, but rather segments the sequence into continuous intervals: output This is a clinically understandable coverage range.

[0036] Step 2.1, Amplitude Weighting: Allow the reliability layer to dominate the segmentation; The amplitude is obtained from step 1. Converted into weights .

[0037] The significance of this step is: The segmentation is no longer affected by the noise layer, but its boundary is determined by a stable and event-driven layer.

[0038] Step 2.2, Weighted Circular Discreteness: Characterize the same stage using phase consistency; This invention uses circumferential statistics to describe whether phases are concentrated within a certain segment (concentration = same phase): , in, Place the phase on the unit circle; the phase changes smoothly within the same phase → the vector sum and length are large → Small; phase changes or mixing occur → phase dispersion → big.

[0039] Because of the use The low reliability layer contributes very little, thus hard-coupling reliability into the segmentation objective.

[0040] Step 2.3, Stability Adaptive Penalty Λ: Suppressing excessive segmentation of noise segments; Traditional DP segmentation is commonly used This fixed penalty makes it easy to accidentally cut off the noise segment.

[0041] This invention uses intra-segment average stability Adaptive adjustment of penalty: If a segment is unstable overall, increase the penalty to prevent noise from being mistaken for multiple false stages.

[0042] Step 2.4, output the segmented coverage interval; Obtain N consecutive intervals (Typical 10-14 segments) This is a provable covering structure: each slice belongs to a certain segment.

[0043] Step 3: Irreplaceable Pilot Selection and Transition Band Closed-Loop Frame Interpolation (IRP+CBW-Loop) – Quantifying the Necessity of Keyframes and Automatically Triggering Encryption Overlay Due to Transition Risks The core pain point to be addressed: Traditional representative frame selection is mostly based on similarity / center point, and doctors will ask: Why must we choose this one? Can we choose another one? Step 3.1, Irreplaceability IRP: Defining necessity using the loss of evidence after removal; This invention, for each segment, does not use the closest to the center or the highest score, but instead asks a counterfactual question: If this option is not selected, how much of the phase consistency evidence will be lost for this segment? This defines irreplaceability And select the largest frequency as the dominant frequency. .

[0044] The key effects emphasized by this method are: keyframe selection and evidence of necessity. They share the same origin.

[0045] Step 3.2, Transitional Obfuscation Bandwidth (CBW): Quantify the degree of transitional obfuscation; Calculate the minimum window width near each segment boundary. To ensure that both sides reach a stable threshold: Narrow transition: clear phase switching; Wide transition: Severe stage mixing, prone to missed scans.

[0046] Step 3.3, Closed-loop frame interpolation: CBW is not an explanatory indicator, but a control signal (evidence → action): when Large enough to exceed the threshold Automatically select another event pilot within the transition band. It specifically covers ambiguous transitions, reduces the risk of missed reads, and features adaptive encryption triggered by evidence of transition risks.

[0047] Final set of keyframes (triggered conditions): .

[0048] Step 4: Change-Point Driven Evidence Generation (CPDE) – Output an auditable evidence package; The core pain point to be addressed: the black box.

[0049] The evidence output by this invention includes four categories: Coverage evidence: Each interval Dominant frequency intra-segment stability .

[0050] Consistency evidence: Intra-segment amplitude-weighted consistency evidence for each segment (phase concentration, representativeness).

[0051] Evidence of necessity: Evidence of necessity for each pilot segment (the irreplaceability of the pilot segment).

[0052] Transition Risks and Closed-Loop Records: Each Boundary Whether it was triggered, the reason for the trigger, and the event pilot score.

[0053] It can also output as JSON or DICOM-SR to form an integrated structured audit log.

[0054] In one specific embodiment of the present invention, the parameters are as follows: 1. Environment and Data GPU: A100 / H100 grade; Software: Python 3.9+, PyTorch 2.x, DICOM library; Data: multi-layer thickness, multiple reconstruction kernels, multiple temporal phases.

[0055] 2. Example of a key module (taking abdominal CT as an example, D=300) 2.1 Encoder E: Uses 2.5D input (current slice + two neighboring slices) for output .

[0056] 2.2 MPG Perturbation Generation: Reading Layer Thickness Metadata such as reconstruction kernel type and pixel spacing; exist Resampling should be performed within the specified range. Simulate kernel differences using light sharpening / light smoothing that satisfies the MTF monotonicity constraint; Noise is estimated using the noise figure / dose field and equivalent noise is injected; The window width and position trembled slightly within the abdominal window range.

[0057] set up .

[0058] 2.3, STA: Assume A pseudo-random orthogonal projection is obtained. .

[0059] 2.4, EVB: Set window radius , Solve for the prediction vector And obtained (z).

[0060] 2.5, Amplitude and Phase: Pick % quantiles, and scale normalization of layer thickness T; ; Obtained by PHA and construct .

[0061] 2.6, W-CCPS segmentation: Let... ;use With adaptive penalty We can use dynamic programming to find the change points and obtain... Sections and each section ; Calculate the amplitude-weighted complex geometric midpoint for each segment .

[0062] 2.7, IRP Pilot and Closed-Loop Frame Interpolation: Calculation , get ;calculate ,like (generally If the value is 6), then select within the transition zone. .

[0063] As an event pilot. The final output is approximately... to Several keyframes are triggered.

[0064] 3. Example of Evidence Output Output: Coverage range ; Dominant frequency ; Boundary confusion bandwidth (No frame interpolation triggered); Pilot necessity .

[0065] Explanation: z=155 is the dominant frequency in this stage, and its removal will significantly reduce the evidence of consistency within the segment. (Higher); the transition bandwidth is 4 slices, the transition is relatively clear, and no additional event pilot is required.

[0066] 4. Transition band triggering example (large CBW) Traditional Top-K keyframes tend to focus on locations with the "most information" (such as the center of an organ or the area with the strongest texture), while transition zones (stage transitions, gate regions, and areas near blood vessel bifurcation) may be missed because they lack maximum values.

[0067] This invention uses CBW to clearly quantify the degree of transition ambiguity. When CBW is large, the system automatically adds an event pilot to forcibly cover the areas that doctors are most likely to miss.

[0068] When the system calculates at a certain boundary... ( Figure 2 , Figure 3 (Boundary 1) The width of this transition band is less than the threshold. This indicates that the phase can recover to a stable and concentrated state within a short range, which is a case of clear transition. Therefore, the system only outputs the dominant frequency keyframes for each coverage segment. No additional event pilots are added.

[0069] When the system is calculated at another boundary ( Figure 2 , Figure 3 (Boundary 2) The width of this transition band is greater than the threshold. This indicates that the phase is mixed over a wide range, belonging to a high-risk fuzzy transition situation. At this time, the system automatically... Selecting event pilots within the transition band ( Figure 3 (As indicated by asterisks) to encrypt keyframe coverage in the phase transition area most prone to being missed; and record the triggering reason (CBW exceeding the threshold) and event pilot score in the evidence package.

[0070] This invention provides a CT keyframe selection method based on complex phase constellation coding. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. A CT keyframe selection method based on complex phase constellation coding, characterized in that, Includes the following steps: Step 1: Slice complex phase symbolization: Transform the slice into a complex phase symbol that can be reconstructed; Step 2: Amplitude-coupled circular phase-change dot codebook segmentation W-CCPS, turning frame selection into a verifiable continuous coverage area; Step 3: Irreplaceable pilot selection and transition band closed-loop frame interpolation; Step 4: Variable point driven evidence generation CPDE, outputting an auditable evidence package.

2. The method according to claim 1, characterized in that, Step 1 includes: Step 1.1, Metadata Constraint Perturbation Generation MPG: Transform the enhancements into a protocol-physically consistent perturbation family; Read the DICOM metadata M and construct a perturbation family that is physically consistent with the case protocol. ; Step 1.2, Stability (STA): Verifiable reliability is defined using the consistency definition under the perturbation family; Encode the perturbation slice to obtain the feature vector of the same slice under slight changes compatible with the current protocol. Fingerprint consistency among two or more perturbations on the same slice can be obtained using projection sign consistency. , The fingerprint consistency ratio between perturbation pairs, of which Indicates direct proportion; Step 1.3, Event-based EVB: Use dual anchors to predict rupture and identify stage transitions or structural events; Search within the windows on both sides of the z-axis for the prediction vector that best explains the current slice. To obtain the residual : ； in It is the original slice The embedding vector; Step 1.4, Amplitude Gating (AMG): Combines stable reliability with event-criticality amplitude. ; ； Step 1.5, Phase PHA: Anatomical progress is represented by curvature accumulation. ， in It is the phase of the anatomical process, along the embedded sequence. The cumulative turning angle of the trajectory; It is the angle between adjacent tangent vectors; mod is the modulo function; Step 1.6, obtain the complex symbols for each layer: ， in, It is the complex phase symbolization of the slice, where e represents the natural constant and j represents the imaginary number.

3. The method according to claim 2, characterized in that, Step 2 includes: Step 2.1, obtain the amplitude from Step 1. Converted into weights Where β is the emphasis coefficient of the weight; Step 2.2, weighted circumferential dispersion; Step 2.3, Stability Adaptive Penalty Λ: Using intra-segment average stability Adaptive adjustment of penalties; Step 2.4: Output the segmented coverage interval.

4. The method according to claim 3, characterized in that, Step 2.2 includes: describing whether the phases within the segment are concentrated using circumferential statistics. ， in, It is an interval The amplitude-weighted circumferential dispersion, b and b are two different slices; This means placing the phase on the unit circle.

5. The method according to claim 4, characterized in that, In step 2.4, N consecutive intervals are obtained. , Indicates the first The coverage areas of slices a and b in each segment.

6. The method according to claim 5, characterized in that, Step 3 includes: Step 3.1, Irreplaceability IRP: Defining Irreplaceability And choose the first The one with the greatest irreplaceability within the segment is the dominant frequency. ; Step 3.2, Transitional Obfuscation Bandwidth (CBW): Calculate the minimum window width near each segment boundary. This ensures that both the narrow and wide transition sides reach a stable threshold. At the border Nearby, expand the window width to both sides. Left window right window Calculate whether the phase concentration on the left and right sides exceeds the threshold. The window is considered stable if and only if the weighted circular phase concentration of the defined window is greater than or equal to the threshold. The corresponding minimum window size is... ; Step 3.3, closed-loop frame interpolation: when Large enough to exceed the threshold Automatically select another event pilot within the transition band. , It is the risk threshold for triggering closed-loop frame interpolation. If the transition band is too wide, encrypted sampling is required; if The transition is clear and no frame interpolation is needed.

7. The method according to claim 6, characterized in that, In step 3.3, the final set of keyframes is obtained. : 。 8. The method according to claim 7, characterized in that, In step 4, the following four types of evidence are output: Coverage evidence, including: each interval Dominant frequency intra-segment stability ; Consistency evidence includes: intra-segment amplitude-weighted consistency evidence for each segment; Evidence of necessity, including: evidence of necessity for each pilot segment; Transition risks and closed-loop records, including: each boundary Whether it was triggered, the reason for the trigger, and the event pilot score.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing program code that, when executed by the processor, causes the processor to perform the steps of the method as described in any one of claims 1 to 8.

10. A storage medium, characterized in that, It stores a computer program or instructions that, when run on a computer, perform the steps of the method as described in any one of claims 1 to 8.

Images