Local Illumination Normalization for Robust Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems struggle with inconsistent image processing due to wide variations in illumination, object presence, and texture, especially in uncontrolled lighting conditions, making reliable object detection challenging in applications like robotics and augmented reality.
Innovation Solution
An adaptive local contrast normalization (ALCN) approach using a normalizer CNN trained jointly with a detector CNN to predict normalization parameters for each image window, allowing robust object detection without requiring extensive training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Stability of the object's composition
If traditional normalization techniques are applied to images with varying illumination, then processing consistency may be improved, but noise amplification occurs and robustness deteriorates
Solution Approach 1:
The patent applies parameter changes by using a CNN to dynamically predict normalization parameters (alpha and beta) for each image window based on local illumination characteristics. Instead of using fixed normalization parameters, the system adapts parameters locally to match the specific illumination conditions of each region, thereby maintaining processing consistency without amplifying noise. This is achieved by training the CNN to output parameters that equalize illumination across different regions while preserving local image characteristics.
2Measurement precision
If illumination normalization is applied to improve object detection, then detection accuracy is improved, but device complexity increases due to CNN training and application
Solution Approach 1:
The patent applies segmentation by dividing the image into multiple windows or regions and processing each window independently with the CNN. This allows the system to handle complex illumination variations in different parts of the image separately, improving detection accuracy while managing computational complexity through parallel processing. The image is segmented into overlapping or non-overlapping windows, and normalization parameters are predicted for each window individually.
Solution Approach 2:
The patent applies preliminary action by pre-training the CNN model on a large dataset of images with varying illumination conditions before deployment. This preliminary training phase allows the system to learn optimal normalization strategies for different illumination scenarios, so that during actual object detection, the pre-trained model can quickly adapt to new images without requiring extensive real-time computation. The joint training of normalization and detection networks further simplifies the deployed system.
3Reliability
If extensive training data is used to train normalization models, then model robustness is improved, but loss of time and computational resources increases
Solution Approach 1:
The patent applies merging by combining the illumination normalization network and the object detection network into a single jointly-trained system. Instead of training separate normalization and detection models independently, the system trains both networks together with a combined loss function, allowing the normalization parameters to be optimized directly for detection performance. This joint training approach achieves robust normalization with less training data and time because the two tasks reinforce each other during training.
Solution Approach 2:
The patent applies self-service by enabling the system to automatically adapt to different illumination conditions without requiring manual parameter tuning or extensive annotated training data for each specific scenario. The CNN learns to self-adjust normalization parameters based on the input image characteristics, and the joint training framework allows the system to self-optimize both normalization and detection capabilities simultaneously during a single training phase.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for normalizing an image by an electronic device is described. The method includes obtaining an image including a target object. The method also includes determining a set of windows of the image. The method further includes, for each window of the set of windows of the image, predicting parameters of an illumination normalization model adapted to the window using a first convolutional neural network (CNN), and applying the illumination normalization model to the window to produce a normalized window.