Color Classification by Hue and Similarity Without Deep Learning

The patent segments the color classification process into two stages: first determining an initial category based on hue dimension, then selecting sub-color categories as candidates, and finally determining the target category through similarity comparison. This segmentation reduces the classification space and eliminates the need for extensive deep learning training while maintaining accuracy.

The patent changes the approach from using large-scale deep learning models to using color space parameters (hue, saturation, value) and similarity calculations. By transforming the classification problem into parameter-based comparison, it achieves accurate color classification without the high training costs associated with deep learning.

By segmenting the classification into initial category determination and target category selection, the patent reduces the time required for model training and inference. The hue-based initial classification quickly narrows down candidates, and similarity comparison provides rapid final classification without extensive computational time.

The patent performs preliminary classification by determining the initial category based on hue dimension before proceeding to target category selection. This preliminary action reduces the search space and eliminates the need for time-consuming deep learning training processes.

Segmenting the classification process reduces computational resource consumption by dividing the task into manageable stages with smaller data sets at each stage, eliminating the need for resource-intensive deep learning training.

The patent replaces expensive deep learning models with simpler, computationally lighter color space parameter comparisons and similarity calculations, achieving the same classification accuracy with significantly reduced resource consumption.