Depth Map Generation Using Image Block Variance Analysis

The patent creates a reference image as a copy of the input image, then compares corresponding blocks between the input image and reference image to calculate variance magnitudes. This copying approach enables depth estimation without training by leveraging the statistical properties of image blocks and their variances, thus resolving the contradiction between obtaining accurate depth information and avoiding time-consuming training processes

The patent replaces the conventional classification-based mechanical processing system with a statistical variance analysis system. Instead of using trained classifiers to estimate depth, the system substitutes this with a mathematical approach that calculates variance magnitudes of image blocks and uses these variances to determine depth values, eliminating the need for training while maintaining depth estimation capability

The patent divides the input image into multiple image blocks and compares each block with its corresponding block in the reference image. By segmenting the image into smaller units and analyzing variance magnitudes of individual blocks, the system achieves accurate depth estimation from a single input image without requiring multiple images or complex training processes, thus maintaining both operational simplicity and depth accuracy

The patent transforms the depth estimation problem from a classification-based approach requiring trained parameters to a variance-based approach using statistical parameters. By changing from using trained classification models to using variance magnitudes of image blocks as the key parameter, the system achieves accurate depth information from a single input image with simplified operation and reduced cost