Developing learning based intelligent fusion for deblurring confocal microscopic images Nabeela Kausar, Abdul Majid n , Syed Gibran Javed Biomedical Informatics Research Lab, Department of Computer and Information Sciences, Pakistan Institute of Engineering and Applied Sciences, Nilore 45650, Islamabad, Pakistan article info Article history: Received 7 December 2015 Received in revised form 28 June 2016 Accepted 8 August 2016 Keywords: Confocal microscopic frames Learning based intelligent models Deblurring Multi-focus Image fusion abstract The demand of high quality confocal microscopic images is increasing for critical tasks such as study of living tissues at cellular resolution and disease diagnosis. The results of such tasks are often affected by the blur introduced in microscopic images. Removal of blur from multi-focus microscopic images without deteriorating their visual quality is a challenging task. The confocal microscopic images are obtained by averaging their frames. This process introduces the blurring artefacts that degrade the quality of mi- croscopic images. In this work, we presented learning-based intelligent fusion to minimize the blurring artefacts of confocal microscopic images. The quality of these images is improved using the proposed individual and ensemble fusion models. In the proposed scheme, block-based features are extracted from the blurred images. These informative features are then used to develop the individual models that construct the fused images using their fusion maps. The predicted information of the individual fusion maps is then combined to construct the ensemble-based fused image. The proposed learning-based approach has demonstrated improved quantitative and qualitative results compared to rule-based fusion approaches. The proposed fusion models can be employed as a useful tool in confocal microscopy frames to generate the improved quality images with reduced blurring artefacts. & 2016 Elsevier Ltd. All rights reserved. 1. Introduction With the increase of new emerging biomedical applications, the requirement of high quality all-in-focus microscopic images has in- creased. These microscopic images play very vital role in under- standing the behavior of various micro-organisms. For example, to visualize the germination of fungal spores for rust detection in wheat plants, study of living tissues at cellular resolution, view where specific molecules of interest are localized, and quantitative assay system for cellular dynamics (Huang, 1981; Rodighiero et al., 2015; Roelfs et al., 1992). These microscopic images help in the in- vestigation to cure from different diseases i.e. the wheat-rust causes a significant reduction in the overall production of wheat every year. The adverse effect of this disease and its cure can be found through an imaging application based on confocal microscopy. The images generated through the confocal microscopy are degraded with diffractive blurring. These images are used for the extraction of data at cellular resolution in the studies of biological systems. However, the use of this type of data acquisition technique becomes limited due to the introduction of diffractive blurring in the captured images. As a result the image pixels cor- respond to each object point is overlapped with the neighboring pixels. The deconvolution and averaging techniques are commonly employed to deblur the microscopic images. The deconvolution based technique is developed by subtracting or reassigning the out-of-focus pixels. The most annoying artifacts, due to deconvo- lution, are the apparent loss of dim-features or the “blowing up” of very bright ones (Wallace et al., 2001). On the other hand, the averaging is applied to the confocal microscopic frames to obtain an improved confocal image. However, simple averaging process introduces smoothness in the resultant image, which causes the loss of useful information in the form of lines and edges. Therefore, to minimize this diffractive blurring effect, fusion approach can be employed in the microscopic images i.e., the frames of the confocal microscopy images can be fused using an efficient fusion ap- proach. The frames fusion step may lead to the sharper image resulting in the reduction of image data error. Sometimes, to ob- tain the good quality images of living tissues, the operator has to adjust several parameters of the microscope. The proposed fusion approach may be helpful to save the critical time of the operator/ technician consumed in the precise adjustment of parameters. In this scenario, we have developed learning-based intelligent mod- els for deblurring confocal microscopic images. By employing the Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/engappai Engineering Applications of Artificial Intelligence http://dx.doi.org/10.1016/j.engappai.2016.08.006 0952-1976/& 2016 Elsevier Ltd. All rights reserved. n Corresponding author. E-mail addresses: nabeela.kausar@pieas.edu.pk (N. Kausar), abdulmajiid@pieas.edu.pk (A. Majid), gibranjaved_11@pieas.edu.pk (S.G. Javed). Engineering Applications of Artificial Intelligence 55 (2016) 339–352