Tests show 100 dpi PDF downsampling harms small Chinese OCR, while English holds up

·1 views

A tester downsampled a 300 dpi A4 scan containing Chinese and English text at several resolutions and measured OCR accuracy. Settings at 150 dpi and above kept accuracy at 99% or higher for both languages while cutting file size to about a quarter of the original. At 100 dpi, 6 pt Chinese text fell to 81.3% accuracy, with some characters misread, whereas English remained largely legible. At 72 dpi, Chinese text was mostly unreadable by OCR, though the file shrank only modestly beyond the 100 dpi version. The tester suspects Chinese characters' many strokes cannot be separated at small pixel heights, though this remains unconfirmed.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Integrated Knowledge Representation Project Outlines a Reusable Logic Inference Engine

This chapter from the STATERA Library's Volume 02 on Knowledge Representation presents an integrated project for building a logic inference engine. The engine derives new knowledge from explicitly stated facts and logical rules, aiming to turn formal reasoning principles into executable operations for symbolic AI. Its architecture comprises a knowledge base, a logical expression parser, an inference controller, working memory, and a query interface, with an emphasis on correctness, transparency, modularity, and extensibility. The material covers forward and backward chaining, unification, and resolution-based refutation as core reasoning mechanisms.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Neuro-Symbolic AI Combines Neural Learning with Symbolic Reasoning

Neuro-symbolic artificial intelligence is an approach that integrates neural network learning with symbolic knowledge representation and logical reasoning. It aims to pair neural networks' ability to recognize patterns in complex data with symbolic systems' capacity to express concepts, relationships, constraints, and rules. The approach emerged from the historical split between symbolic AI, which emphasized explicit logic, and connectionist AI, which focused on learning distributed representations from data. Integration is challenging mainly because neural outputs are continuous while symbolic systems operate on discrete entities, so uncertainty management and semantic alignment are needed. Strategies include using neural models to generate symbolic facts for reasoning engines, and embedding symbolic constraints into training, with differentiable reasoning and knowledge graphs playing key roles.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Rule-Based Systems in Knowledge Representation: Chapter 04 Overview

This chapter from the STATERA Library's Volume 02 on Knowledge Representation introduces rule-based systems as a symbolic AI approach that lets computers reach decisions using explicitly defined knowledge and logical conditions. Its core principle is separating the knowledge base, which stores IF-THEN rules, from the inference engine that determines which rules apply to a given set of facts. The text explains forward and backward chaining as the two basic reasoning methods and traces the historical development from expert systems such as MYCIN to modern uses in business process automation and compliance. It highlights interpretability and consistency as key strengths, while noting limitations including the cost of maintaining large rule sets and difficulty handling uncertainty and changing conditions.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Mars' Valles Marineris Found Still Tectonically Active, Study Reports

Researchers at Columbia's Lamont-Doherty Earth Observatory report that Valles Marineris, a roughly 4,000-kilometer canyon system splitting Mars, shows ongoing tectonic movement. The finding challenges the view that the feature is geologically inactive. The study's coverage on Hacker News sparked discussion that shifted from planetary geology to the long-held belief that amblyopia can only be treated in early childhood. The article's author, who was diagnosed with amblyopia as a child, describes using a virtual reality app that presents separate visual inputs to each eye as a training approach.

0 comments Read more at DEV Community