Filtering-out poor-quality images for data preparation

2026-07-14 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

An approach is proposed that filters out poor-quality images for various environmental and camera impacts, rather than denoising them. This method addresses limitations of traditional CNN-based denoising techniques, which often require specific training datasets, image resizing, and are limited to certain types of environmental and camera noise. The new approach assesses image quality using a specific metric and applies an optimum threshold to remove low-quality images, while ensuring a sufficient quantity remains for deep learning model development. Demonstrated on real and simulated traffic and object recognition data, the approach achieved an average recognition accuracy of 93.8% for traffic sign recognition and 84.9% for object recognition, outperforming existing methods.

Key takeaway

For Machine Learning Engineers preparing image datasets for critical applications like autonomous vehicles, consider implementing a quality filtering step instead of traditional denoising. Your models can achieve higher recognition accuracy, as demonstrated by 93.8% for traffic signs and 84.9% for object recognition, by proactively removing poor-quality images based on an assessment metric and threshold. This ensures robust data quality for deep learning model development.

Key insights

Filtering out poor-quality images using an assessment metric and threshold surpasses traditional denoising for diverse noise types.

Principles

• Image quality assessment can replace denoising.
• Thresholding effectively removes poor-quality data.
• Maintain sufficient data for DL model training.

Method

Assess image quality using a metric, then apply an optimum threshold to filter out poor-quality images. Ensure enough data remains for deep learning model development.

In practice

• Implement quality filtering for autonomous vehicle datasets.
• Apply threshold-based filtering to object recognition data.

Topics

• Image Quality Assessment
• Data Preparation
• Deep Learning
• Object Recognition
• Traffic Sign Recognition
• Autonomous Vehicles

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• Efficient, Validation-Free Intrinsic Quality Estimation for Large-Scale Face Recognition Datasets — Intrinsic Quality (IQ) offers a rapid, validation-free method to assess face recognition dataset potential before full training.
• Flow Map Denoisers: Traversing the Distortion-Perception Plane for Inverse Problems — Flow Map Denoisers use a single parameter to continuously balance distortion and perception in image restoration and inverse problems.
• Rethinking Infrastructure Inspection as Image Difference Classification: A Traffic Sign Case Study — Image difference classification with instruction-based models reduces data dependency for infrastructure defect detection.
• From level set evolution to threshold optimization: A grayscale level set framework for image segmentation — Explicit geometric regularization is unnecessary for image segmentation if images are sufficiently smooth.
• Assessing VLM Reliability for Medical Image Quality Evaluation Under Corruption and Bias — VLMs for medical image quality assessment exhibit significant vulnerabilities to image corruption and contextual metadata bias.
• Tool-IQA: Augmenting Image Quality Assessment with Simple Tools — Tool-IQA enhances VLM-based image quality assessment by integrating dynamic inspection tools like Magnifier and Gamma Corrector.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.