Micro-image definition evaluation using multi-scale decomposition and gradient absolute value
As an important instrument to observe the micro world, optical microscope has been widely used in medical health, biological detection, industrial production and other related fields. Because the depth of field of the microscope imaging system is very small, clear images can only be acquired within a small range of depth variations. The evaluation and determination of microscopic image sharpness directly affect the precision of microscope autofocus and become an important index to measure the imaging quality of microscopic system. With the development of multimedia technology and digital image, the requirement of automation for micro-instrument and equipment has been gradually improved, and more and more attention has been paid to the micro-image sharpness evaluation algorithm based on image processing, which is of great significance for the rapid and accurate microscopic autofocus and imaging system performance evaluation.
In the process of obtaining micro-image, it is easy to be affected by the noise of imaging link, and the traditional definition evaluation methods often have multiple false peaks, which have their own limitations in the aspects of algorithm sensitivity, unbiasedness and unimodal.
The research team of Cui Guangmang, Ningbo Yongxin Optics Co., Ltd. and School of Electronics and Information, Hangzhou Dianzi University, proposed a micro-image sharpness evaluation algorithm by combining the design of multi-scale decomposition tool and gradient absolute value operator, aiming at the quality evaluation of micro-image autofocus and imaging system. Non-subsampled shearlet transform (NSST) was used to decompose image edge information into detail layers of different scales. Through the setting of anti-noise threshold, gradient absolute value energy operators of high frequency and low frequency were calculated. By taking advantage of the changing characteristics of high and low frequency energy information, effective microscopic definition evaluation was realized.
NSST is a very effective multi-scale image decomposition tool proposed in recent years. It has simple mathematical structure, parabolic scale, stronger directional sensitivity and optimal sparsity and other characteristics. At the same time, NSST can avoid the sampling process for the loss of image information, which has the advantage of translation invariance and stability. The contour of image, edge, texture details, can be better expressed. It can be suitable for the extraction of image features and provide more information on judgment for clarity evaluation algorithm. The decomposition process flow diagram is shown in Fig.1.
Fig. 1 Flowchart of non-subsampled Shearlet transform
The frequency domain subband coefficient gradient absolute value energy and operator (Sum of absolute gradient, SAG) were used to construct the sharpness evaluation algorithm, which replaced the square operator in the energy gradient function and reduced the computational complexity and improved the computational efficiency while representing the image edge sharpness. In addition, the anti-noise threshold value is introduced, and only the frequency sub-band values larger than the threshold value are calculated when the evaluation function value is calculated, which enhances the anti-noise performance of the algorithm.
In this paper, the frame flow chart of microscopic image sharpness evaluation combining multi-scale decomposition and gradient absolute value operator is shown in Fig. 2. NSST was used to decompose the input micro-image to be evaluated at multiple scales, and one low-frequency sub-band image and several high-frequency sub-band images in different directions at different scales were obtained. The gradient absolute value energy and operator of each sub-band image are calculated, and the noise interference to the evaluation result is eliminated by setting the anti-noise threshold, then the weighted high-frequency sub-band energy and operator and low-frequency sub-band energy and operator are obtained, and finally the clarity evaluation operator is obtained by the ratio of the two.
Fig.2 Micro-image definition evaluation method using multi-scale decomposition and gradient absolute value
Simulation and real experiments were carried out to obtain micro-image sequences with different sharpness. Several comparison algorithms are used to calculate the definition index, and the normalized evaluation result curve is obtained. The normalized evaluation result curve of the micro-image sequence of noise in the simulation experiment is shown in Fig. 3. It can be seen from the analysis that the normalized evaluation curve basically reflects the clarity change trend of the image sequence, but the anti-noise ability of the Tenengrad function and LS evaluation algorithm is poor, and the evaluation curve shows obvious fluctuation and forms obvious secondary peak, which affects the accuracy of the algorithm. The algorithm proposed in this paper has strong anti-noise ability and is basically free from noise interference, showing good stability and robustness.
Fig.3 Normalized evaluation value curves for noisy microscopic image sequences in simulation experiments
The research team of Cui Guangmang, Ningbo Yongxin Optics Co., Ltd. and School of Electronics and Information, Hangzhou Dianzi University, are dedicated to the research and application promotion of computational imaging, intelligent image processing and sensor technology. The knowledge system of optics, electronics, algorithms, machinery and other disciplines are considered and comprehensively applied to solve practical engineering problems. The main research directions include optoelectronic imaging and intelligent image processing technology, computational imaging theory and method, optoelectronic detection technology and so on. The research team includes 5 teachers and more than 10 postgraduate students. At present, the team have published more than 40 SCI/EI scientific research papers and applied for more than 40 national invention patents, with more than 10 patents been authorized. During the cooperation between the associate professor Zhao Jufeng, a member of the research team, and Ningbo Yongxin Optics Co., Ltd., we have won one second-class prize of Ningbo Science and Technology Award in 2017 and one second-class prize of Zhejiang Province Science and Technology Progress Award in 2018, based on the project of "Research and Application of Microscopic Image Quality Improvement Technology".
Cui Guangmang, Zhang Keqi, Mao Lei, et al. Micro-image definition evaluation using multi-scale decomposition and gradient absolute value[J]. Opto-Electronic Engineering, 2019, 46(6): 180531.