Image Coding TRIZ Cases

A computer-implemented method compresses deep image data by combining pixel samples based on object identifiers and depth variations.

Sub-patches separate geometry from attributes in V3C mesh compression, resolving sparse connectivity handling limits.

Encoding map data with relative vertex offsets reduces storage capacity and processing latency while maintaining visual fidelity of complex building structures.

A picture data transmission method recodes images to reduce file size while maintaining quality.

A deep learning-based in-loop filter detects reference regions to combine with current frames.

Packing decomposed volumetric video patches into 2D planes reduces data volume while maintaining 6DOF rendering quality.

A decoder parses atlas sub-bitstreams containing displacement values and identifiers to reconstruct specific mesh regions.

Generative adversarial networks localize sound sources on video pixels using optical flow data.

An iterative compression method updates entropy models after each cycle to control data loss within acceptable bounds.

A point-cloud decoding device configures a maximum initial node size parameter to control recursive tree division processes.

A decoding apparatus signals bit depth information in a sequence parameter set to reconstruct attribute data accurately.

A learned image compression model encodes visual data into compact representations optimized for downstream machine learning tasks.

A compression system subdivides image data into partitions to optimize color palettes and reduce file sizes.

A 3D data encoding device converts point cloud geometry into occupancy maps via an N-ary tree and generates a bitstream using a correspondence table.

Dynamic mesh segmentation separates attribute and geometry patch scaling to resolve temporal compression performance deterioration in volumetric video.

Parallel GPU kernels process compressed image packets to exceed 300 MB/sec, resolving sequential bottlenecks.

A multi-level nested set representation groups non-zero frequency coefficients into hierarchical structures for efficient entropy encoding.

Host device generates concurrent backlight configuration data with pixel data to eliminate lag artifacts in head-mounted displays.

A prediction network generates video blocks using intra and inter models.

Explicit weighted prediction modifies motion compensation blocks using specific flags and indices to enhance inter prediction accuracy.

LERC limits per-pixel error during compression to reduce file sizes while preserving elevation data precision.

A hybrid neural network encoder processes source data and side information to generate guided output for video reconstruction.

A point cloud transmission device segments data into independent patches for selective geometry-based or video-based compression.

Replacing point cloud data with geometric functions reduces transfer time while maintaining shape accuracy.

A gradient-driven compression technique generates DNN-specific quantization tables to reduce image size for edge inference.

Neural networks predict user gaze to drive adaptive video rendering and compression strategies.

Signaling neural network post-filter parameters via syntax elements enables advanced video processing within coding frameworks.

Grouping feature map channels enables spatial and channel context extraction to resolve low encoding rates in traditional image compression.

Adaptive point cloud attribute coding selects candidate points near a centroid to generate optimized levels of detail layers.

A dynamic reduction function maps neighborhood configurations to tree leaf nodes for entropy coder selection in point cloud encoding.

Compressing bounding volume hierarchies by encoding node inheritance structures minimizes memory consumption while maintaining ray-tracing accuracy.

A dynamic mesh encoder segments vertex coordinates into 3D sub-blocks and maps them to 2D patches for efficient compression.

A geometry encoder uses a machine learning regressor to select optimal compression parameters for 3D data.

A polygon mesh generation method scans image lines to build and simplify rectangles enclosing visible pixels.

A point cloud encoding method selects predictor points based on relative geographic position to enhance attribute prediction accuracy.

Local geometry projection compresses dynamic point clouds via 2D image mapping, resolving the trade-off between transmission efficiency and geometric accuracy.

A predictive coding method determines sub-volume occupancy patterns to guide entropy context selection for point cloud data.

A 3D point cloud encoding method selects candidate points using Morton codes to calculate prediction residuals for efficient bitstream generation.

A tile change list buffer tracks modified image regions to enable efficient frame buffer updates.

Geometry point cloud compression encoder avoids single occupancy value coding when inter prediction or planar mask data exists for current octree nodes.

A three-dimensional point generation method rotates or inverts nearby points to create converted data for resolution enhancement.

An encoder uses a machine learning model to select an encoding intra-prediction mode, reducing computational complexity while maintaining coding efficiency.

A 3D mesh compression method projects divided patches onto a base mesh to encode geometry and texture data.

Merging blocks with similar distribution patterns suppresses encoding efficiency loss caused by noise-induced Octree pattern variations in natural images.

A 3D data encoding method embeds prediction tree control information into bitstreams to facilitate efficient decoding.

Neural networks encode captured images into compact spatio-temporal polynomial latent spaces for efficient 3D scene reconstruction.

Applying an adaptive smoothing filter to occupancy map edges reduces transmission bitrate while preserving reconstruction quality and minimizing artifacts.