A $50,000 Camera you Already Own

Hyperspectral Recovery from RGB Images

A $50,000 Camera you Already Own

Conventional cameras capture images using only three frequency bands (red, blue, green), while the full visual spectrum is a much richer representation that facilitates a wide range of additional and important applications. A new technology allows conventional cameras to increase their spectral resolution, capturing information over a wide range of wavelengths without the need for specialized equipment or controlled lighting.

Figure 1 | HC-Vision?s hyperspectral reconstruction methods: A rich hyperspectral prior is collected, a corresponding hyperspectral dictionary is produced and projected to RGB. Once produced, the dictionary may be used to reconstruct novel images without a additional hyperspectral input. (Bild: Hc Vision)

Figure 1 | HC-Vision’s hyperspectral reconstruction methods: A rich hyperspectral prior is collected, a corresponding hyperspectral dictionary is produced and projected to RGB. Once produced, the dictionary may be used to reconstruct novel images without a additional hyperspectral input. (Bild: Hc Vision)

Hyperspectral (HS) imaging systems are capable of collecting the complete spectral signature reflected from each point in a given scene, producing a much more spectrally detailed image than that provided by RGB cameras. Although these systems are widely used in industrial and scientific settings, they have yet to find a place in the consumer market due to their cost, size, and slow acquisition process (often requiring close to one minute to acquire a single image). HC-Vision’s story began at the Ben-Gurion University Interdisciplinary Computational Vision Laboratory where the company’s co-founders, Prof. Ohad Ben-Shahar and Boaz Arad, were studying the properties of natural hyperspectral images. To this end they began collecting what is now the largest, most detailed natural hyperspectral image collection published to date. Analysis of this database revealed that, in the case of natural images people experience in typical indoor and outdoor environments, underlying hyperspectral information can be accurately recovered from its RGB projection.

Figure 2: Filter testing platform used to design single chip systems. (Bild: Hc Vision)

Figure 2: Filter testing platform used to design single chip systems. (Bild: Hc Vision)

Hyperspectral images from RGB

Unlike the three channel images that RGB cameras (and human eyes) produce, hyperspectral images contain dozens or hundreds (and from an abstract theoretical point of view, infinite number) of wavelength bands. But even simpler hyperspectral cameras contain at least 31 channels, a number representing the division of the visible spectrum to spectral bands of 10nm, and so even in this case, the 31-to-3 dimensionality reduction that occurs while projecting hyperspectral information to RGB appears rather severe,. And yet, HC-Vision’s methodology is already able to recover the former from the latter with 90 percent to 95 percent accuracy over a wide variety of scenes. In addition to producing state-of-the-art results at the time of publication, this approach produced comparable results to previous methodologies which relied on hybrid HS/RGB input. This methodology often surpassed the performance of the latter (in both accuracy and computation time) despite a significant information disadvantage.

Figure 3: With the specially crafted filters, the quantum efficiency of existing sensors can doubled without compromising color accuracy. (Bild: Hc Vision)

Figure 3: With the specially crafted filters, the quantum efficiency of existing sensors can doubled without compromising color accuracy. (Bild: Hc Vision)

Hyperspectral dictionary

The key to the success of the new methodology lies in leveraging a rich hyperspectral prior – embodied by their extensive hyperspectral database, as well as the sparsity of hyperspectral information in natural scenes. This is achieved via a unique, patented, dictionary learning approach: First, several terabytes of hyperspectral data are reduced to a dictionary less than 1MB in size. Then, the dictionary is adapted to the target RGB camera. Finally – the adapted dictionary can be used to recover hyperspectral information from previously unseen images taken with the target RGB camera. This method produces high accuracy results in a general setting, but accuracy can be increased even further by tailoring the hyperspectral prior to a specific application. For example: if one intends to use the system in an agricultural setting, its dictionary may be trained exclusively on agricultural images, thus increasing its accuracy in that setting. The prospect of low cost, snapshot hyperspectral imaging from a handheld device is indeed quite enticing, but HC-Vision’s technology has exciting applications for conventional imaging as well. Since the reconstruction process does not explicitly require a camera with an RGB-like response, optical designs previously unsuited for conventional imaging can now be used without compromising color accuracy. Indeed, over the past year, the company have been developing a camera system with optical filters optimized for both quantum efficiency and hyperspectral reconstruction. On average, this system is twice as light-sensitive as a comparable RGB camera. In addition to improved sensitivity, the system also increases hyperspectral estimation accuracy by over 30 percent. While the testing platform at figure 2 relies on four cameras, the final product will be produced in a single-chip, single-lens configuration using the same manufacturing techniques used for conventional RGB sensors. While these systems are still in active development and not yet available for purchase, they are steadily progressing towards a market-ready version. In addition to their independent imaging platform, HC-Vision are currently negotiating with several industry-leading companies looking to incorporate the new technology into their upcoming products.

More Links

Sparse Recovery of Hyperspectral Signal from Natural RGB Images -https://goo.gl/48UV5Q

ICVL natural hyperspectral image database – http://icvl.cs.bgu.ac.il/hyperspectral

Interdisciplinary Computational Vision Laboratory – http://icvl.bgu.ac.il

Ben-Gurion University of the Negev -www.bgu.ac.il

 

Hc Vision

Das könnte Sie auch Interessieren