Crate oximedia_scene 

Scene understanding and AI-powered video analysis for OxiMedia.

oximedia-scene provides comprehensive scene understanding and intelligent video analysis capabilities for the OxiMedia multimedia framework. This includes:

• Scene Classification: Classify scenes (indoor/outdoor, day/night, landscape, portrait)
• Object Detection: Lightweight patent-free object detection
• Activity Recognition: Recognize activities (walking, running, sports)
• Shot Composition: Analyze framing (rule of thirds, symmetry, leading lines)
• Semantic Segmentation: Segment image into semantic regions (sky, ground, people)
• Saliency Detection: Identify visually important regions
• Aesthetic Scoring: Rate aesthetic quality of frames
• Event Detection: Detect events in sports and live content
• Face Detection: Lightweight face detection (Haar cascades)
• Logo Detection: Detect brand logos and graphics

Patent-Free Algorithms

All algorithms are carefully selected to be patent-free:

• HOG (Histogram of Oriented Gradients): Object detection
• Haar Cascades: Face detection
• Color Histograms: Scene classification
• Motion Histograms: Activity recognition
• Spectral Saliency: Attention prediction
• Graph-based Segmentation: Semantic regions
• Rule-based Composition: Framing analysis

Modules

• classify: Scene, content, and quality classification
• detect: Object, face, logo, and text detection
• activity: Activity and sports recognition
• composition: Composition rules, balance, and depth analysis
• segment: Semantic and foreground/background segmentation
• saliency: Saliency detection and attention prediction
• aesthetic: Aesthetic quality scoring and feature extraction
• event: Event detection for sports and live content
• features: Feature extraction and descriptors

Example

use oximedia_scene::classify::scene::SceneClassifier;
use oximedia_scene::detect::face::FaceDetector;
use oximedia_scene::composition::rules::CompositionAnalyzer;

// Example usage
let classifier = SceneClassifier::new();
let face_detector = FaceDetector::new();
let composition = CompositionAnalyzer::new();

Re-exports

pub use error::SceneError;

pub use error::SceneResult;

Modules

action_beat

Action beat detection and analysis for video scenes.

activity

Activity recognition from motion patterns.

adaptive_scene

Adaptive scene detection with complexity histogram analysis.

aesthetic

Aesthetic quality assessment.

audio_visual_correlation

Audio-visual correlation module for detecting synchronisation between audio events and visual changes.

camera_motion

Camera motion estimation and classification.

classification

Scene classification: genre detection, mood estimation, and location tags.

classify

Scene and content classification.

color_temperature

Scene color temperature estimation and analysis.

common

Common types and utilities used across modules.

complexity_detector

Complexity histogram-based scene detection.

composition

Shot composition analysis.

content_moderation

Content moderation analysis for video frames.

continuity_check

Continuity checking for video scene sequences.

crowd_density

Crowd density estimation for video frames.

depth_of_field

Depth-of-field computation for cinematographic scene analysis.

detect

Object, face, logo, and text detection.

emotion_recognition

Emotion recognition module for analyzing facial expressions.

error

Error types for scene understanding.

event

Event detection in video.

face_landmark

Facial landmark detection and geometric face analysis.

features

Feature extraction and descriptors.

lighting_analysis

Lighting analysis for video frames.

location

Location and setting analysis for scenes.

mood

Scene mood analysis.

motion_energy

Motion energy module for quantifying overall motion intensity per scene segment.

object_tracker

Multi-object tracking with Kalman filter state estimation.

pacing

Scene pacing analysis.

saliency

Saliency detection and attention prediction.

scene_boundary

Scene boundary detection: types, descriptors, and frame-based detector.

scene_captioning

Scene captioning module for generating natural-language descriptions from scene features.

scene_graph

Scene graph structure for hierarchical narrative representation.

scene_metadata

Scene-level metadata: tags, environmental descriptors, and keyword sets.

scene_score

Scene quality scoring: per-category scores and aggregate scene score.

scene_stats

Per-scene metrics and aggregate statistics across a sequence of scenes.

scene_tags

Automatic hierarchical scene tagging with confidence-scored labels.

segment

Image segmentation — foreground/background and semantic region detection.

segmentation

Scene segmentation: shot boundary detection, scene grouping, and transition types.

shot_type

Shot type classification for cinematographic framing analysis.

storyboard

Scene storyboard generation.

summarization

Scene summarization: key shot extraction and scene-level digests.

temporal_graph

Temporal graph module for connecting scene analysis results across time.

text_detect

Advanced text detection using edge density, stroke width analysis, and text classification.

thumbnail_selector

Thumbnail selector module for choosing the most visually representative frame per scene.

transition

Scene transition analysis.

visual_quality_map

Per-region visual quality mapping.

visual_rhythm

Visual rhythm extraction and analysis.