Impulse Noise Detector Performance Measure Based on Intensity Volume Long Bao 1 & Karen Panetta 1 & Sos Agaian 2 Received: 17 May 2019 /Revised: 8 August 2019 /Accepted: 15 August 2019 # Springer Science+Business Media, LLC, part of Springer Nature 2019 Abstract An accurate detector performance evaluation method provides a fair comparison platform and can also support in parameter optimization for existing Impulse noise detectors in the applications of medical imaging. The Impulse noise detector performance measure (INDPM) package is widely applied as tools for quantitative comparison among detectors, which contains recall measure, accuracy measure, precision measure, specificity measure and F-measure. However, these five measures suffer from limited accuracy in correctly evaluating the performance of a detector and are not in well agreement with human subjective evaluation. To solve this problem, five new measures are proposed by introducing a new concept of intensity volume to form a new Impulse noise detector performance package (IV-INDPM). Using a standard image dataset, we conduct experimental and comparative tests with 32 different original images and 5 different existing detectors. Results demonstrate the superior perfor- mance of each new measure within IV-INDPM in reaching a much closer agreement with human subjective evaluation, com- pared to existing measures in INDPM. Even though five new measures are efficient in evaluating detectors’ performance from different perspectives, a new benchmark algorithm (IND-BA) is proposed as a robust and overall metric for ease of general- purpose use by making the most of these five new measures. Comparison results demonstrate its efficiency and accuracy. Keywords Image denoising . Impulse noise . Detector . Detector performance measure 1 Introduction Medical imaging equipment makes significant contributions to modern medical science by providing visual information of the internals of the human body. Different types of medical imaging machines have been designed for different visual information capture, such as X-ray radiography, magnetic res- onance imaging, medical ultrasonography or ultrasound, pos- itron emission tomography (PET) and single-photo emission computed tomography (SPECT). A clear medical image can support the correct diagnose and subsequent effective treat- ment. However, due to the high radiation property of medical imaging equipment, medical imaging suffers from noise cor- ruption in the visual image generation, especially for nuclear images, MRI, CT and ultrasound imaging [1]. These corrupted noise artifacts will mislead the doctors in their diagnosis and prognosis, which is vital to the life of patients. Hence, the problems of noise corruption in these specific medical imag- ing is still a challenge. The presence of noise can drastically affect image quality, which can manifest itself by impacting the performance of consumer electronics [2, 3]. First, noise negatively impacts the users’ visual perceptions. Second, it limits the perfor- mance of many advanced computer vision systems (e.g. de- tection and recognition). Furthermore, this negative influence might be extended to many advanced image processing sys- tems. Given all the detrimental impacts noise can have on consumer electronics, this paper presents a method to help medical imaging system designers select the best robust noise removal method for the product. Different types of noise have different statistical properties, and specific approaches have been designed for addressing each particular type of noise [4–7]. Among all the types of * Long Bao Long@eecs.tufts.edu Karen Panetta Karen@ece.tufts.edu Sos Agaian Sos.Agaian@csi.cuny.edu 1 Department of Electrical and Computer Engineering, Tufts University, Medford, MA 02155, USA 2 Department of Computer Science, The City University of New York, Staten Island, NY 10314, USA Journal of Signal Processing Systems https://doi.org/10.1007/s11265-019-01475-4