Geometric Image Transformation TRIZ Cases

This MSFA case uses guide and difference images to preserve edges and improve spatial resolution during demosaicing.

Patch-based atlases encode machine-learning regions of interest at suitable resolution without coding the full frame uniformly.

A learned dual back-projection process upsamples sparse point clouds, improving 3D detail while reducing acquisition and storage demands.

This case combines directional downsampling and interpolation to correct chromatic aberration with lower CPU resource use.

Virtual reference frames improve boundary interpolation in scaled-picture video encoding.

This case trains a super-resolution image model with frequency-domain loss to align high-frequency components and improve image sharpness.

A staged back projection network refines residual images to improve super-resolution quality while preserving data fidelity.

This case uses occupancy mapping, prediction, and shape-adaptive transforms to compress point clouds with unoccupied positions.

Parallel pixel block processing and PCA-based joint-color encoding reduce latency while preserving image quality in artificial reality.

Skip outside-fisheye pixels during RAW development to reduce processing load.

This case uses a cross-component resampling inloop filter to reconstruct reduced-resolution chroma with luma guidance.

Reduced-resolution processing runs on the wearable while a companion device completes HDR tone mapping and merging.

Photon-number-resolving cameras capture pixel-level distributions to separate multiple sources without stringent alignment.

Condition-based pixel selection improves interpolation across image areas.

A parameter transformation model replaces standard convolution layers to cut memory and computation while preserving inference accuracy.

This case reconfigures MPM information by projection region to improve 360-degree image compression without uniform complexity growth.

This engineering case adjusts application resolution from designated and non-designated screen areas to expand visible content on small displays.

Adaptive layer transformation reduces inference load while preserving model accuracy.

Horizontal azimuth mapping projects large point clouds onto regular 2D planes, reducing algorithm complexity and projection time.

Spherical latent encoding expands object discovery beyond complex slot-based models.

Combines super-resolution MRI and voxel-wise modeling for accurate tissue characterization.

This case uses perceptual quantization and iterative angular-resolution adjustment to pack MPI patches within target atlas sizes.

Two image sensors and electronic cropping maintain binocular overlap and vertical alignment during magnified 3D surgical viewing.

This case uses CNN filter sets to downscale images before transmission and restore definition through AI upscaling at the terminal.

Segmented CFA blocks and green-rich shells support binning, demosaicing, low-light sensitivity, and improved image SNR.

The case uses segmented parallel coding with a neural model to improve compression speed and limit artifacts at batch divisions.

A display control system links replica movement to object-image orientation and reveals cut-plane information at a virtual plane.

Microfluidic FPM combines wide-field imaging and reconstruction for liquid biosignature detection.

Neural encoders and decoders use optical flow, correction data, and predicted-image features to limit artifacts, bitrate, and quality loss.

A pre-trained GAN optimizes latent vectors against target images to produce realistic, tileable material maps with less manual editing.

Control projection and frame averaging to prevent blur in fast rotating radiography.

Multiband TDI sensors and fiber-bundle beams extend exposure time, improving imaging quality while reducing power density and photodamage.

This case splits image regions across scaler cores, using alignment data to prevent boundary artifacts during parallel scaling.

3D scans and neural networks identify rooms and stories, reducing manual effort and errors in standardized property layouts.

Surface coordinates and logical object links let AR overlays connect virtual objects with physical or virtual props.

Usage-based buffer selection and super-resolution preserve the original frame rate while improving images for larger displays.

Align vector edges precisely with threshold-guided detection and real-time feedback.

Full-frame processing is costly; event-based regions and downscaled images reduce pixels and memory for classification.

Recursive block splitting and projection-specific prediction help decode high-resolution 360-degree images with better compression.

A sparse color filter array combines panchromatic and filtered pixels across burst frames to improve color detail and light sensitivity.

Wavelet-blended document images create customizable training data for real-time object detection in unstructured documents.

Down-sampling, CNN inference, enlargement, and shape-based post-processing accelerate accurate region extraction from radiographed images.

Automatic reference positions and separated timeline periods reduce manual steps when extracting regions from VR frame sequences.

A dual-frequency image model uses reference and input features to produce sharp images at continuously adjustable resolutions.

Non-uniform atlas packing improves 6DoF video quality allocation and decoding efficiency.

This case uses a device’s imaging hardware to capture displayed calibration images and improve embedded machine learning reliability.

This case splits full-frame pixels into partial fields for parallel ray tracing and filtering, reducing processor data transfers.

This case uses estimated depth, point clouds, and forward warping to approach NeRF-quality views at 140–1000× higher speed.

This case applies target gain values to feature maps, enabling multiple compression bit rates within one image encoding model.

This case uses encoder-decoder networks to predict synthetic frames, reducing hitching and tearing while maintaining rendering quality.