Abstract
Although recent studies on object recognition using deep neural networks have reported remarkable performance, they have usually assumed that adequate object size and image resolution are available, which may not be guaranteed in real applications. This paper proposes a framework for recognizing objects in very low resolution images through the collaborative learning of two deep neural networks: image enhancement network and object recognition network. The proposed image enhancement network attempts to enhance extremely low resolution images into sharper and more informative images with the use of collaborative learning signals from the object recognition network. The object recognition network with trained weights for high resolution images actively participates in the learning of the image enhancement network. It also utilizes the output from the image enhancement network as augmented learning data to boost its recognition performance on very low resolution objects. Through experiments on various low resolution image benchmark datasets, we verified that the proposed method can improve the image reconstruction and classification performance.
Highlights
Object recognition is one of the well-conquered problems in machine learning owing to the use of deep learning techniques [1]–[7]
The object recognition network is based on an existing well-trained model, and we propose systematic retraining strategies for this network that utilize the ability of the pre-trained network efficiently and augment its recognition performance on low resolution images
Using the training signals originating from the object recognition network, the image enhancement network (IEN) can generate images with improved quality in terms of appearance and perception
Summary
Object recognition is one of the well-conquered problems in machine learning owing to the use of deep learning techniques [1]–[7]. Performance considering that the top-5 error rate of humans is 5.1% [12] Despite these prominent results, the focus on lowresolution object recognition has been weaker than that on high-resolution images. The average resolution of the images used in ILSVRC is 482 × 415 pixels [13] Those images contain backgrounds and multiple objects, they can retain sufficient information about each object, which enabled deep networks to extract rich visual features from them and achieve notable classification performance. There is no guarantee that the deep networks designed for high resolution object recognition can perform well when classifying extremely low resolution images, in which much of the useful object-related information is collapsed
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have