As modern displays continue to increase in resolution, means of capturing images and videos at such high resolutions can be prohibitively expensive. This is especially true in the infrared domain. Image super-resolution, or upsampling, has often been applied to improve the aforementioned problem. Deep learning models have been proposed to reconstruct high quality high-resolution images from a low-resolution base. Previous solutions require a massive number of parameters which necessitate a large amount of free memory and computation power or they fall apart when applied to the infrared domain. As a result, many modern super-resolution models are not entirely practical. One difficult aspect in IR super-resolution is that IR images are inherently noisy, causing a poor signal-to-noise ratio, due to characteristics of IR sensors and internal reflections within the lenses. Because of this, super-resolution in IR must also act somewhat as a denoiser. Therefore, we propose a highly efficient, super-resolution model capable of producing single-image super-resolution in the IR domain.
The objective of image compression is to reduce irrelevance and redundancy of the image data to be able to store or transmit data in an efficient manner by minimizing the number of bits required to represent an image accurately. JPEG is capable of achieving an image compression ratio of 10:1 with little perceptible loss in image quality using standard metrics, and has become the most widely used standard image compression in the world since its release. Traditionally, compression techniques have relied on linear transforms to approximate 2-D signals (images), and the omission of specific constituent vectors has been mostly arbitrary. These techniques can save incredible amounts of memory while retaining image integrity. Recently techniques have been developed that use neural networks to approximate these signals. These networks offer the advantage of decorrelating image data to find a series of vectors to represent an image that is smaller than traditional techniques by estimating gradient descent, thus finding the minimum number of bits required to represent an image. Expansion to the development of these architectures is happening rapidly through informed design drawing upon other fields that have recently seen increased focus such as computer vision and image analysis applications. A novel efficient neural network is proposed in this work to compress infrared images at state of the art levels while preserving overall image quality to handle the demands spanning from the daily commute to combat environments.
Modern displays are steadily increasing in resolution, though sensors can be prohibitively expensive to capture images and video at such high resolutions. Image super-resolution, or upsampling, has recently been applied to alleviate these shortcomings. There exist many deep learning image super-resolution models that reconstruct very high quality high-resolution images from a low-resolution base. However, most of these models use a tremendous amount of parameters, requiring a large amount of free memory and computational power to super resolve a single image. As a result, many modern super-resolution models are not entirely practical due to the computational or memory usage requirements. We propose a highly efficient, small super-resolution model utilizing the sub-pixel convolution block for single image super-resolution.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.