Back to search

Rochester Institute of Technology

Content Aware Intra and Inter Prediction Using Texture and Motion Analysis

Abstract

dc:description.abstract

Modern video codecs rely on highly optimized block based intra and inter prediction frameworks, in which block partitioning is determined through exhaustive rate distortion optimization (RDO) searches. These exhaustive searches evaluate many possible block configurations to identify the partitioning that minimizes overall rate distortion cost, and they have proven extremely effective across diverse visual content. This thesis explores a complementary perspective by investigating whether block partitioning can be guided more directly by content characteristics specifically, texture complexity for intra prediction and motion complexity or motion density for inter prediction. For intra prediction, we introduce a texture adaptive approach that utilizes multi-scale Gabor energy analysis to characterize local spatial structure. The intuition is that regions with more intricate or high-frequency textures benefit from smaller block sizes, while smoother areas can be represented efficiently with larger blocks. This content-aware partitioning is designed to work alongside, rather than replace, traditional RDO, providing an informed initial structure that can reduce reliance on exhaustive searches. For intra prediction, we propose a motion adaptive block selection strategy driven by block level motion segmentation. Areas with complex or dense motion are assigned to smaller block sizes, enabling more precise motion compensation, while temporally stable regions favor larger partitions. Across a variety of test sequences, the proposed framework demonstrates encouraging results, achieving an average 1.1 dB PSNR improvement over the baseline reference configuration in controlled experiments. These findings suggest that intelligently guiding block partitioning using content complexity measures may offer a promising complementary direction for future research in video coding.

Degree

thesis:*
Name thesis:degree_name
Electrical Engineering (MS)
Level thesis:degree_level
Thesis
Discipline thesis:degree_discipline
Electrical and Microelectronic Engineering, Department of

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Danda, Yashaswi Karthik Reddy
Contributors dc:contributor
  • Eli Saber

Subjects

dc:subject × 4

Identifiers

dc:identifier.*
Repository record dc:identifier
https://repository.rit.edu/theses/12470

Chain of custody

source
Harvested from
Rochester Institute of Technology
Base URL
repository.rit.edu/do/oai/
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Danda, Yashaswi Karthik Reddy. Content Aware Intra and Inter Prediction Using Texture and Motion Analysis. Thesis thesis, https://repository.rit.edu/theses/12470