SEM Image Quality Assessment Based on Intuitive Morphology and Deep Semantic Features

Abstract

The widespread use of scanning electron microscopy (SEM) has increased the requirements for SEM image quality. SEM images obtained by electron beam feedback have more complex texture features than natural images obtained by optical imaging, and this condition results in poor performance of algorithms used for assessing natural image quality on SEM datasets,meanwhile,the field of SEM image quality assessment(IQA) is mostly aimed at specific distortion types. In order to solve the above two problems,to address the rich texture, few edges, and extreme sensitivity to the distortion degree of SEM images, we propose a texture and semantic IQA (TSIQA) method for SEM images based on sparse mask and information entropy increase. First, we construct a neural network containing sparse mask module (SMM), which is used to extract intuitive texture features in the spatial and channel domains. Simultaneously, information growth attention (IGA) is introduced into SMM to detect the difference between current and past features of the network for extracting deep semantic information. The quality assessment experiments on SEM image datasets show that compared with the state-of-the-art IQA methods, including popular noreference techniques adapted to the SEM-IQA, the TSIQA has superiority in typical criteria.

Image quality evaluation for FIB‐SEM images

“…[22][23][24][25] For individual SEM images, recent no-reference methods 26,27 evaluate blurring combining gradients and grey value statistics. Also, Wang et al 28 employ a neural network trained on 650 high-and low-quality SEM images of ants, metal, stamens, colloids and minerals (details described in Li et al 29 ) to classify SEM images into 'good' and 'bad' quality. Their focus is on rich texture images of separate objects which is in contrast to the homogeneous material samples studied here.…”

“…For individual SEM images, recent no‐reference methods 26,27 evaluate blurring combining gradients and grey value statistics. Also, Wang et al 28 . employ a neural network trained on 650 high‐ and low‐quality SEM images of ants, metal, stamens, colloids and minerals (details described in Li et al 29 …”