Entropy-Based Objective Functions for Uniform, Discriminative Image Features

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image feature extraction networks suffer from nonuniform image distribution and poor discrimination due to non-uniform image features.

Innovation Solution

Perform normalization processing on vectors corresponding to each pixel in a target feature map set to generate a target vector set, determine hash coding for each vector, calculate prior probabilities of these hash codings, and generate a target function based on entropy of these probabilities.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If existing image feature extraction networks are used, then image features can be extracted, but the image distribution is nonuniform and discrimination is poor

Engineering Contradiction:

Improveimage feature discriminationVSAvoidimage distribution uniformity

Core Design Contradiction:

Manufacturing precisionVSStability of the object's composition

Solution Approach 1:

The patent transforms image features by converting pixel vectors into hash coding representations, fundamentally changing the parameter space from continuous vector values to discrete hash codes. This parameter transformation enables uniform distribution across hash buckets and improves discrimination by creating distinct categorical representations for different image features

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent introduces hash coding as an intermediary representation between raw pixel vectors and final image features. This intermediary layer with uniform distribution properties acts as a mediator that improves both the uniformity and discrimination of the extracted image features while maintaining the essential information

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS12437502B2Method for generating objective function, apparatus, electronic device and computer readable medium

Publication Date: 2025.10.07 DOUYIN VISION CO LTD

AI summary

A method for generating a target function is provided. The method includes: performing normalization processing on a vector corresponding to each pixel in a target feature map set to generate a target vector, so as to obtain a target vector set; generating hash coding corresponding to each vector in the target vector set, to obtain a hash coding set; determining a prior probability of each hash coding in the hash coding set; and generating a target function based on an entropy of the prior probability.