A high-efficiency acquisition method of LED-multispectral images based on frequency-division modulation and RGB camera

Highlights

• LED-multispectral imaging is one efficient spectral image acquisition technique.

• RGB camera based LEDMSI can acquire 3-band images in a single exposure.

• Frequency-division modulation can achieve the simultaneous acquisition.

• A high-efficiency acquisition method of LED-multispectral images is proposed.

Abstract

LED illumination based multispectral imaging (LEDMSI) is one of the promising techniques of fast and effective spectral image acquisition. RGB camera based LEDMSI can acquire 3-band images in a single exposure, which has practical, high-efficiency and low-cost advantages. Frequency-division modulation technology can achieve the simultaneous acquisition of multi-wavelength images. In this paper, a high-efficiency acquisition method of LED-multispectral images based on frequency-division modulation and RGB camera is proposed. And its effectiveness is verified by the experiment with dual-frequency modulation of six wavelengths (two multiplexed illuminations) as example. In the experiment, two multiplexed illuminations modulated by two carrier frequencies are used as the light source to acquire an image sequence, and grayscale images are obtained by “R/G/B” 3-channel separation. Further, the Fast Fourier Transform is used to demodulate the time series of corresponding pixels of each grayscale image, the images of each multiplexed illumination are obtained, respectively. In the result and discussion section, the multispectral images obtained by the method in this paper is compared with the multispectral images obtained without combining frequency-division modulation. Through the result, it can be concluded that the method of combining frequency-division modulation with RGB camera is a high-efficiency acquisition method of high-quality multispectral images, which provides a reference for the LED-multispectral imaging technology.

Introduction

Multispectral imaging is typically used to capture spectral reflectance of the imaged scene. And it is not limited to the visual range, but also can be applied in ultraviolet, near infrared, and infrared spectrum as well, depending on the response range of the sensor. Through the spectral imaging systems, the spectral reflectance of a scene reflecting the unique property of an object can be recovered from the acquired images. Therefore, multispectral imaging has a wide range of applications, such as remote sensing [1], biometric security [2], [3], medical imaging [4], [5], culture and heritage [6] etc.

There are already many different types of multispectral imaging techniques and systems. Among them, one promising technique of multispectral imaging, which is of the primary interest in this paper, is based on Light Emitting Diode (LED) illumination [7], [8], [9], [10], [11]. In recent years, LED illumination based multispectral imaging (LEDMSI) has attracted much attention due to its rapid computer-controlled switching capability, robustness, and cost-effectiveness. Availability of high-intensity LEDs with many different colors and peak wavelengths spanning the whole visual range and even infrared region makes it possible to construct more effective multispectral systems. LEDMSI has been used in several applications like underwater geological and biological surveys [12], medical imaging [13] and film scanner [14]. The systems can be classified into two major categories according to the type of camera: monochrome camera based [7], [8], and RGB camera based [10], [11]. In a monochrome camera based LEDMSI (Mono-LEDMSI) system, a group of n different types of LEDs are selected, each type of LED is illuminated in a sequence, and a monochrome camera captures images under the illuminated LED, thereby generating n-band multispectral images. In an RGB camera based LEDMSI (RGB-LEDMSI) system, combinations of LEDs whose number and wavelength might be constrained by the availability and other practical factors of the system are selected. Each combination of LEDs is lit at a time, and a color camera is used to capture 3-band images. A K $=$ 3 $\times$ N-band image of a scene can be acquired with N different combinations of LEDs in N exposures, so the acquisition can be speed up by a factor of three.

For monochrome camera and frequency-division modulation, our research team has proposed a number of imaging methods. Frequency-division modulation can not only suppress interference, but also can achieve the acquisition of multi-wavelength images simultaneously, saving image acquisition time. In 2018, X. Yang et al. proposed an optimized illumination method, in which a shaped-function signal was applied to increase the dynamic range of LED-multispectral imaging system [15]. In 2019, H. Li et al. proposed and demonstrated the multi-wavelength “synergy effect” in LED-multispectral images obtained by frequency-division modulation [7]. In the same year, F. Liu et al. proposed a method combining modulation–demodulation–frame-accumulation technique and pattern recognition to realize heterogeneity classification [16], in which the modulation–demodulation–frame-accumulation technique was adopted in pre-processing experiment to enhance the image signal. In the above research, a monochrome camera was used as the image acquisition terminal, and the modulation and demodulation were reasonably used in the process to achieve different application purposes.

In addition, for RGB camera and multiplexed illumination (a combination of two or more LEDs in each exposure), some researchers have proposed their systems. Rather than sequentially activating the sources, J. Park et al. proposed a novel technique to find the optimal multiplexing sequence of spectral sources in 2004 [10], so as to minimize the number of acquired images. In 2012, M. Parmara et al. developed a multispectral imaging system that captures multiple acquisitions during a rapid sequence of LED with different colors [17]. In 2015, R. Shrestha et al. proposed a fast multispectral imaging using LED illumination and an RGB camera along with a novel LED selection method [18]. In the same year, R. Shrestha et al. proposed a system which used an RGB camera along with two or three combinations of three different types of LEDs so as to acquire multispectral images of six or nine channels [11]. In 2018, T. Heimpold proposed a new time efficient selection method for the design process of an illumination. It chose an optimal LED combination from an existing database to match a predefined spectral power distribution [19]. From the above, we can get that the researchers are particularly concerned about finding the optimal multiplexed illumination of LEDs in RGB-LEDMSI systems. The determined multiplexed illuminations are activated sequentially, and the corresponding images are collected, then a multispectral video of the scene is obtained.

In the past research work, some have combined frequency-division modulation to improve image quality and increase the dynamic range of LED-multispectral imaging system, and some have proposed multiplexed illuminations in RGB-LEDMSI to speed up image acquisition, but no one has tried to combine the two. When the two are combined, the entire image acquisition process is equivalent to be performed on two dimensions: (1) the color channel of the receiving device itself; (2) frequency-division modulation of multi-wavelength images. Therefore, a high-efficiency method for obtaining multispectral images can be proposed, which may provide a reference for improving the quality of tissue multispectral transmission imaging in the future.