Digital holographic microscopes (DHMs) have been widely applied in material and biological applications. For instance, among many biological and biomedical applications [1–4], DHM systems are used for analyzing cells and tissues [4–8], as well as disease diagnosis and screening [4, 9–11] and assessing the polarimetric properties of biomedical samples[12, 13]. In material science, DHM systems have been used to measure MEMs [14–19], detect and characterize defects [20], as well as to characterize surface topography [21–23]. The performance of DHM technologies relies heavily on computational reconstruction processing to provide trustworthy sample information. The required computational reconstruction algorithms are uniquely dependent on the optical configuration of the DHM system. An incorrect selection of the reconstruction algorithms leads to distorted and inaccurate amplitude and phase measurements. DHM systems record the interference pattern (e.g., hologram) generated between the scattered light from the sample, named object wavefront, and a known reference wavefront. DHM systems operate in off-axis, slightly off-axis, or in-line (also known as on-axis) configurations based on the interference angle. Therefore, the selection of the DHM reconstruction algorithms depends on this interference angle. For example, off-axis DHM systems enable the reconstruction of the complex amplitude distribution of an object wavefront from a single hologram since the three components of the hologram are entirely separable in the Fourier domain [24, 25]. Therefore, the reconstruction algorithm in off-axis DHM systems requires the spatial filtering of the sample frequencies from the hologram’s spectrum. In contrast, the spectral components of a hologram in the Fourier domain overlay totally or partially in in-line and slightly off-axis DHM systems, respectively, requiring the acquisition of multiple phase-shifted holograms and the application of phase-shifting (PS) techniques [1, 26] to reconstruct the desired object information. In addition, the selection of the DHM reconstruction algorithms also depends on whether the DHM imaging system operates in telecentric or non-telecentric regimes. DHM systems operating in the telecentric regime only require the interference angle compensation in the off-axis and slightly off-axis configuration. Oppositely, non-telecentric DHM systems should compensate for the spherical phase factor recognized in DHM and associated with a non-telecentric imaging system [27–30]. Finally, the DHM configuration can operate in an image-plane configuration, meaning that in-focus DHM holograms are recorded, so there is no need to apply numerical propagations to focus the amplitude and phase images. However, if out-of-focus holograms are recorded, the user should numerically propagate the complex amplitude distribution to provide in-focus images. Among the different numerical propagation algorithms to reconstruct DHM images, the most used computational approaches in DHM are the angular spectrum for short propagation distances [31, 32] and the Fresnel transform algorithm for large propagation distances [33].

Still, each research group within the DHM community develops and implements its own computational reconstruction algorithms based on its experimental digital holography (DH) and DHM systems. Nonetheless, some research groups have developed and implemented libraries and plugins to address the need for an open-source reconstruction toolbox in DHM and DH. The main difference between DHM and DH systems is the utilization of a microscopic imaging system in the object wave. For example, in 2010, Shimobaba et al. developed a numerical library, e.g., the GWO library, for diffraction calculations using a Graphics Processing Unit (GPU) [34]. This GWO library enabled the numerical propagation of complex amplitude distributions using the angular spectrum and Fresnel approaches. Since the GWO library was based on the C language and, consequently, was not user-friendly, in 2012, the same authors developed a new C++ class library for diffraction and calculations using computer-generated holograms (CGHs) [35]. In 2015, Piedrahita-Quintero et al. developed a JAVA plugin for numerical wavefields propagation [36]. This plugin enabled the propagation of complex amplitude distributions using angular spectrum, Fresnel, and Fresnel–Bluestein approaches. The pluggin’s most important feature is that it is embedded within the well-known open-source software for image processing named ImageJ. Piedrahita-Quintero et al. validated their plugging by numerically propagating experimental holograms recorded in off-axis DH and DHM systems. In 2017, the same authors upgraded their previously-developed JAVA-based plugin to a GPU-accelerated library, JDiffraction [37]. The OpenHolo library, developed by Hong et al. in 2020, can generate holograms using the most popular CGH algorithms [38]. This library also includes standard tools for holography, like phase unwrapping algorithms and reconstruction algorithms using the Rayleigh-Sommerfeld diffraction integral and the numerical Fresnel propagation algorithm. That same year, Trujillo et al. developed an ImageJ-based open-source plugin for in-line digital lensless holographic microscopy (DLHM) [39]. This plugin contains two modules to simulate holograms using the discrete version of the Rayleigh–Somerfield diffraction formula and to reconstruct holograms using the Kirchhoff–Helmholtz diffraction integral. More recently, the Manoharan Lab at Harvard University has implemented the HoloPy library, a Python-based library, to perform scattering and optical propagation theories [40]. Like the JAVA plugin from Trujillo et al. [39], the HoloPy library focuses on in-line DLHM systems.

Despite all these efforts, a library containing the needed computational reconstruction approaches to reconstruct DHM images, regardless of the optical configuration of the system, does not still exist. This paper presents a Python library focused on DHM applications, named pyDHM, for reconstructing DHM images for various experimental DHM implementations. We aim to provide the DH and DHM community with a complete set of tools for holographic processing in a widely supported and easy-to-use programming language. The library consists of four packages. The first package contains a set of useful functions, such as calculating the hologram spectrum and displaying the amplitude and phase images. The second package is related to reconstructing in-line and slightly off-axis DHM systems using phase-shifting techniques. The third package reconstructs phase images from off-axis DHM holograms using telecentric and non-telecentric configurations. Finally, the last package includes algorithms for propagating the complex amplitude distribution using the angular spectrum and the Fresnel and Fresnel-Bluestein transform approaches. Our proposed library has been validated with experimental and simulated holograms recorded using different setups.

DHM systems are based on optical interferometry, which involves the digital recording of the interference pattern (e.g., digital hologram) between the complex amplitude distribution scattered by a microscopic object (e.g., the object wavefront) and a uniform reference wavefront. DHM systems are colloquially known as hybrid imaging systems since the final reconstructed amplitude or phase images are obtained after further computational processing of the digital hologram. This second stage, known as the reconstruction stage, depends on the optical configuration of the DHM system. This section thoroughly describes different DHM configurations and their corresponding numerical reconstruction algorithms.

For simplicity, we have chosen a traditional DHM setup based on a Mach-Zehnder interferometer operating in transmission mode, see Fig 1(A). The light generated from a laser of wavelength λ is collimated by a converging lens (CL) and the resultant plane wavefront impinges a cubic beam splitter (BS1) to generate the object (O) and the reference (R) wavefronts. A microscopic imaging system is inserted in the object arm of the DHM system. This microscopic imaging system comprises an infinity-corrected microscopic objective (MO) lens and a tube lens (TL) which forms an image of the wavefront scattered by the microscopic sample with amplitude distribution o (x,y). If the object is set at the front-focal plane (FFP) of the MO lens, the image of the sample whose amplitude distribution is u_IP (x,y) is then obtained at the back-focal plane (BFP) of the TL. This plane is known as the microscope’s image plane (IP). The complex amplitude distribution u_IP (x,y) produced by the microscope at the IP is given by uIP(x,y)=1M2eik(2fMO+d+fTL)exp(ik2C(x2+y2))×[o(xM,yM)⊗2P(xλfTL,yλfTL)], (1) where k = 2π/λ is the illumination wavenumber, ⊗₂ denotes the 2D convolution operator, f_MO is the focal length of the MO lens, f_TL is the focal length of the TL, d is the distance between the BFP of the MO and TL lenses (Fig 1), and M = —f_TL / f_MO stands for the lateral magnification of the imaging system which does not depend on the distance d. P(u,v) is the 2D Fourier transform of the amplitude transmittance of the pupil distribution, p(x,y). The quadratic phase factor exp(ik2C(x2+y2)) with a radius of curvature C=fTL2fTL−d is associated with a non-telecentric geometry (d ≠ f_TL) for the optical microscope. When the microscope operates in the telecentric regime (d = f_TL), no quadratic phase factor appears in Eq (1).

The hologram, captured at any plane from the IP, is the result of the interference between the complex wavefield produced by the microscope at a distance z from the IP, u(x,y;z)=iλzeikz{uIP(x,y)⊗2exp(ik2z(x2+y2))}, (2) and a tilted plane wavefront, r(x,y)=IRexp[ik(sinθx·x+sinθy·y)], (3) where I_R is the irradiance of the reference wavefront, and θ = (θ_x,θ_y) is the vector representation of the titled reference angle to the optical axis, which coincides with the center of the object wavefront. In Mach-Zehnder-based DHM systems, this angle can be changed by tilting the optical elements that reflect the reference wavefront R (e.g., the BS2 and/or M2 in Fig 1(A)). The irradiance distribution of the hologram h(x,y;z) is h(x,y;z)=|u(x,y;z)|2+|r(x,y)|2+u(x,y;z)r*(x,y)+u*(x,y;z)r(x,y), (4) where z < 0 refers to planes located in front of the IP, |·|² represents the square module, and the symbol * stands for the complex conjugate operation. In Eq (4), the first two terms are related to the irradiance of both object and reference wavefronts. On the other hand, the third and fourth terms in Eq (4) encode the complex amplitude information of the object wavefront scattered by the sample. These terms represent the object’s real and twin images. From Eq (4), it is clear that the object information o(x,y), encoded in u(x,y;z) is mixed with other undesired terms. The reconstruction algorithms aim to separate these undesired terms from the object information to provide well-contrast amplitude and phase images from the complex in-focus amplitude distribution u_IP (x,y) = u(x,y;z = 0) with minimum distortions. Assuming that the reference wavefront is plane, the 2D Fourier transform of the hologram H(u,v;z) is H(u,v;z)=DC(u,v;z)+U(u−sinθxλ,v−sinθyλ;z)+U*(u+sinθxλ,v+sinθyλ;z), (5) where (u,v) are the transverse spatial frequencies, DC(u,v;z)=U*2U*+R*2R* being *₂ the 2D cross-correlation operator. The capital letters refer to the 2D Fourier transform distributions to simplify our notation. The spatial frequencies of the DC term are always placed at the center of the hologram spectrum. However, the frequencies of the ±1 terms, respectively the U(·) and U*(·) terms, are located symmetrically around the DC term at locations that depend on the interference angle θ = (θ_x,θ_y). In other words, the different components of the hologram spectrum may overlap based on the interference angle between the two wavefronts of the DHM system. Knowledge of the hologram’s spectral composition is critical to selecting the reconstruction method. For example, off-axis DHM systems are those in which the angle between the object and reference wavefronts is such that the terms in Eq (5) are not superimposed [Fig 1(D)]. The reconstruction method for off-axis DHM systems involves a spatial filtering approach using a single hologram [24, 25]. The other extreme case occurs when no interference fringes are observed in the hologram since the interference angle between wavefronts is zero. Therefore, the three terms in Eq (5) entirely overlap [Fig 1(B)]. These DHM systems operate in in-line (or on-axis) regime. The third DHM configuration, slightly off-axis DHM systems, lies between these two extremes; the interference angle between both wavefronts is not null but small enough to produce some overlapping between the different components of the hologram spectrum [Fig 1(C)]. For in-line and slight off-axis DHM systems, the reconstruction algorithms involve phase-shifting (PS) techniques, requiring the recording of multiple holograms in which the phase of the reference wavefront is shifted (e.g., phase-shifted holograms) [1, 26]. The main advantage of PS algorithms is that the reconstructed phase images are obtained via point-wise subtractions and division operations between the recorded phase-shifted holograms. Traditionally, PS algorithms are aimed exclusively at strictly in-line DHM systems. We highlight the traditional PS algorithms requiring five, four, and three phase-shifted holograms among the different PS algorithms. In the five- and four-step algorithms, the phase shift between the holograms is π/2. Consequently, the point-wise phase images are reconstructed by φ(x,y)=tan−1(2[h(x,y;3π/2)−h(x,y;π/2)]2h(x,y;π)−h(x,y;0)−h(x,y;2π)), (6) and φ(x,y)=tan−1(h(x,y;3π/2)−h(x,y;π/2)h(x,y;π)−h(x,y;0)), (7) for the five- and four-step algorithms, respectively. The third variable of the hologram distribution in these equations refers to the phase shift of the reference wavefront. For example, h(x,y;3π/2) is a recorded hologram in which there is a phase shift of 3π/2 to the first hologram. One can reconstruct the phase distribution using a three-step PS algorithm with a phase shift of 2π/3 between holograms as φ(x,y)=tan−1(3h(x,y;π/3)−h(x,y;5π/3)h(x,y;5π/3)+h(x,y;π/3)−2h(x,y;π)). (8)

Although the three-step PS algorithm requires fewer holograms, being more suitable for real-time DHM imaging, this implementation is more sensitive to noise than the four- and five-step PS algorithms in experimental conditions. For this reason, four- and five-step algorithms are still used in many phase-shifting DHM configurations.

Due to the experimental difficulty in perfectly aligning the object and reference wavefronts, thus achieving a strictly in-line setup, slightly off-axis DHM systems are commonly preferred. For slightly off-axis DHM systems, traditional PS algorithms must be modified to compensate for the interference angle between both interfering wavefronts. De Nicola et al. proposed a four-step PS strategy with a phase shift of π/2 between consecutive holograms [41]. In this quadrature method, the complex amplitude distribution of the object can be reconstructed by summing the individual products between the recorded holograms and their corresponding digital reference wavefronts u^(x,y)=h(x,y;0)r^(x,y;0)+h(x,y;π/2)r^(x,y;π/2)+h(x,y;π)r^(x,y;π)+h(x,y;3π/2)r^(x,y;3π/2). (9)

Eq (9) provides the complex amplitude distribution of the specimen without the DC term and the conjugate image, enabling the computing of amplitude and phase images via |u^(x,y)| or atan(imag[u^(x,y)],real[u^(x,y)]), respectively.

The main disadvantage of traditional in-line and slightly off-axis PS approaches is that most of these methods require i) accurate knowledge of the phase shift between the recorded holograms, and ii) this phase shift must be equal within the acquisition sequence. These two requirements can be experimentally challenging due to inaccuracies prevalent in most phase-shifting devices, particularly those based on mechanical movements. Several blind PS strategies have been proposed to have multiple holograms with a random and unknown phase shift [42–49]. Among these approaches, we have proposed two blind iterative PS-DHM algorithms for slightly off-axis DHM systems, which are computationally efficient, user-friendly, and robust under noisy conditions [50, 51]. These two blind PS-DHM methods are based on the demodulation of the terms composing the Fourier spectrum of the hologram. In particular, the hologram distribution can be understood as a linear combination between three unknown components {d₀, d₊₁ and d_-1} where the weighting of each component depends on the phase shift: h = d₀ + e^-jΔθ d₊₁ + e^jΔθ d_-1 [50] where d₀ is the two first terms of Eq (4), d+1=u(x,y;z)e−jksinθ⋅x, and d−1=u*(x,y;z)ejksinθ⋅x, being x = (x,y) the vector representation of the lateral coordinates. Based on this composition of the hologram distribution in terms of {d₀, d_+1, and d_-1}, the object information can be reconstructed from either the d₊₁ or d_-1 components using three recorded holograms, {h₁, h₂, h₃} with different phase shifts {Δθ₁, Δθ₂, Δθ₃}. Conversely, the hologram can also be written as the sum of two components{d₀, d₃}: h = d₀+e^iΔθ d₃, where d₃ = d₊₁+ e^-i2Δθd_-1, requiring only two holograms with a phase-shift difference of Δθ to estimate d₃ [51]. An accurate estimation of the d₊₁ or d₃ components involves that the phase shifts used to demodulate the d₊₁ or d₃ components coincide to the ones of the experimental reference wave. When the values of the phase shifts match the real ones, the spectrum of the d₀ component is composed of a unique order for the 2- and 3-step blind PS algorithms. The spectrum of the d₊₁ component is also composed of a unique order for the 3-step blind PS algorithm. Our blind PS-DHM approaches take advantage of this observation, e.g., the expected/real composition of the spectrum of the decomposed terms, to estimate jointly the phase steps and the d₊₁ or d₃ component by minimizing a cost function that quantifies the difference between the absolute value of the Fourier component of the expected/real and residual peaks. Since both algorithms are used for slightly off-axis DHM systems, one must compensate for the linear phase term introduced by the tilted plane reference wave. In addition, the 2-step blind PS algorithm requires the spatial filtering of the spectral frequencies of the d₊₁ term from the spectrum of the demodulated d₃ term. Consequently, the 2-step blind PS approach only works for slightly off-axis DHM systems in which the spectra of the d₊₁ and d_-1 terms do not overlap in the Fourier domain among them. The major limitation of these approaches is that they only work for DHM systems operating in telecentric regime since the center of the ±1 terms in the Fourier transform should correspond to a maximum value.

The reconstruction algorithms in in-line and slightly off-axis DHM systems using PS strategies require multiple recorded holograms, restricting the use of those systems for live-cell imaging and dynamic analysis. Therefore, DHM systems operating in an off-axis regime are the most used DHM system for real-time imaging. These systems allow the reconstruction of an object’s amplitude and phase information from a single recorded hologram. The reconstruction algorithms for reconstructing a phase image in off-axis DHM systems involve two steps. The first step is the spatial filtering of the frequencies related to the object from the hologram spectrum. Due to the off-axis configuration, the ±1 diffraction orders are arranged symmetrically around the DC term in the Fourier space [Fig 1(D)]. From the hologram spectrum [Eq (5)], the spectral object information (i.e., the +1 term) can be filtered [24].

Eq (10) represents the filtered hologram spectrum, which is the spectrum of the sample displaced at the spatial frequencies (sinθ_x/λ, sinθ_y/λ). The amplitude distribution scattered by the sample can be obtained as the absolute value of the inverse Fourier transform of Eq (10). Whereas the spatial filtering step is sufficient for amplitude imaging, quantitative phase imaging requires the phase compensation of the tilting angle between the object and reference wavefronts (the second step of the reconstruction procedure). The phase compensation of the interference angle can be performed in the space domain by multiplying the inverse Fourier transform of Eq (10), h_F(x) and a digital replica of the reference wavefront, r_D(x) as u^(x)=rD(x)·hF(x). Assuming a uniform plane reference wavefront, the generation of a digital reference wavefront requires the knowledge of the interference angle θ = (θ_x,θ_y), which depends on the wavelength of the light source, the features of the digital sensor (i.e., M×N square pixels with pixel size Δ_xy), and the subtraction between the pixel locations of the DC and the +1 terms in the hologram spectrum [52]. Among these parameters, the only one that is unknown a-priori is the location of the +1 term. The determination of this parameter must be precisely executed to provide phase images without phase nuisances. Although this step can be performed manually by generating a set of reconstructed phase images with different phase compensation until the one without sawtooth fringes is obtained, several automated DHM reconstruction approaches [52–56] have been proposed to alleviate this task. Among these works, the pyDHM library implements the following three algorithms that are restricted to off-axis DHM systems operating in telecentric regime. The first two algorithms have been coined the full region of interest (ROI) search [53] and efficient ROI search [56] from Trujillo’s laboratory. The third approach has been investigated and developed by Doblas’ laboratory [52]. The main difference between these three algorithms relies on how the search for the optimal parameters of the digital reference wavefront is performed. In Ref. [53], this search is carried out using nested loops in which different spatial frequencies around the point of maximum energy of the +1 diffraction order (region of interest) are used to generate different compensating angles. Then, different digital reference wavefronts are built with these angles to produce different compensated phase maps of the object. The parameters of the reference wavefront yielding the best-compensated phase image are the proposal’s output. A quantification of the number of phase discontinuities, performed via a summation-and-thresholding metric, is employed to determine the best-compensated phase image in Ref. [53]. More efficiently, in Ref. [56], the search involves a heuristic determination of the spatial frequencies inside the region of interest most likely to provide properly compensated phase maps. The latter path-oriented strategy uses only the resulting best-compensated phase map in each fewer-step nested loop until the same metric to quantify phase discontinuities as in Ref. [53] can not be further improved. Finally, in Ref. [52], the search of the parameters of the digital reference wavefront is based on an unconstrained nonlinear optimization procedure involving the minimization of the inverse of the summation-and-thresholding metric as a function of spatial frequencies around the point of maximum energy of the +1 diffraction order.

In addition to accurately determining the interfering angle between the object and reference wavefronts, the reconstruction algorithms depend on the microscope’s optical configuration. The shape of the ±1 diffraction orders changes between DHM systems operating in telecentric and non-telecentric configurations, depending on the radius of the curvature of the spherical wavefront in Eq (1), C^-1. The smaller the radius of curvature C, the wider the ±1 orders, being the area of the ±1 diffraction orders inversely proportional to C [57]. Fig 2 illustrates the changes in the size of the ±1 diffraction orders based on the radius of curvature C. The ±1 diffraction orders are rectangular-based compact support functions in non-telecentric DHM systems. However, these terms are circular compact support functions in telecentric-based DHM systems whose diameter is related to the resolution of the microscopic imaging system [57], u_c = NA/(λM) where NA is the numerical aperture of the MO lens. The reconstructed phase image from a non-telecentric DHM system is distorted by the quadratic phase factor that appears in Eq (1). Therefore, this quadratic phase factor should be suppressed to reconstruct accurate quantitative phase images in non-telencetric DHM systems. This quadratic phase factor can be suppressed computationally by a point-wise subtraction of the reconstructed phase with and without a sample, e.g., performing a double-exposure technique where two holograms should be recorded [58]. Single-shot computational approaches have also been proposed to eliminate the quadratic phase factor by multiplying the inverse Fourier transform of Eq (10), h_F(x), with the conjugated replica of the distorted phase term, exp[−i2πλC[(x−xC)2+(y−yC)2]] knowing its center (x_C, y_C) and radius of curvature (C). These parameters can be estimated by analyzing the hologram’s spectrum [59].

Finally, numerical propagators of complex wavefields are required in DHM if digital holograms are not recorded at the IP since the object’s reconstructed information must be numerically focused considering the axial distance z between the sensor/hologram plane and the in-focus plane [60]. Conventional numerical propagators are based on the angular spectrum or Fresnel Transform approaches [31, 32]. The angular spectrum approach represents a complex amplitude wavefront as a combination of infinite plane wavefronts following Huygens’ principle. Therefore, the in-focus complex amplitude distribution can be estimated as u^IP(x,y)=FT−1[U^(u,v;z)•exp(−i2πλz1−λ2(u2+v2))], (11) where FT^-1[·] denotes the inverse 2D Fourier transform operator, and U^(x,y;z) is the 2D Fourier transform of the reconstructed complex amplitude distribution [u^(x,y;z)] after applying PS algorithms or spatial filtering and phase compensation the interfering angle. The phase map of the u^(x,y;z) distribution should have been compensated for any quadratic phase factor. The Fresnel Transform approach is another method to solve the Fresnel-Kirchhoff diffraction equation using the paraxial approximation [32]. The paraxial approximation involves that the sensor/hologram plane dimensions are smaller than the propagation distance, i.e., large propagation distances. Based on the Fresnel Transform, the relationship between the out-of-focus distribution, u^(x,y;z), and the in-focus complex amplitude distribution, u^IP(x,y), is expressed as u^IP(x,y)=−i2λ∬u^(x0,y0;z)exp(−iπλz(x2+y2))exp(i2πz(xx0+yy0))dx0dy0. (12)

Some constant phase factors have been neglected in Eq (12). The third term within the integral in Eq (12) is related to the kernel of a 2D Fourier transform with frequencies u = x/z and v = y/z. Although the low-computational complexity of the Fresnel transform enables fast numerical processing, this traditional approach imposes a fixed magnification of the propagated wavefield according to the illumination wavelength and the system’s geometry. A third numerical propagator can be employed to overcome this limitation. In a modified version of the Fresnel approach, the kernel of the Fourier transform in Eq (12) is modified by inserting the Bluestein substitution [33, 61] to convert this expression into a convolution operation in which the magnification of the propagated wavefield can be chosen at will at the expense of higher computational complexity.

In summary, the DHM numerical processing involves computational strategies dependent on the employed optical configuration. In the proposed pyDHM library, we have implemented functions for: i) spatial filtering off-axis holograms, ii) computing complex amplitudes distributions from phase-shifted interferograms in in-line and slightly off-axis setups, iii) compensating the interference angle between interfering wavefronts in off-axis setups, iv) compensating the spherical wavefront of non-telecentric recordings, and v) the propagating complex wavefields to focus non-IP holographic reconstructions.

The pyDHM library consists of four packages. The first utility package includes essential functions such as reading, displaying images, and filtering Fourier spectrums of holograms. The second package is related to reconstructing in-line and slightly off-axis DHM systems using phase-shifting techniques. The third package reconstructs phase images from off-axis telecentric-based DHM holograms. Finally, the last package includes algorithms for propagating complex amplitude distributions using the angular spectrum and the Fresnel and Fresnel-Bluestein transform approaches. This section explains each package in detail, including the functions and required parameters. We also present sample codes and results illustrating the performance of the pyDHM library.

This work presents the pyDHM library, a Python library for the numerical processing of digital holograms registered in DHM systems. The library contains different computational implementations for: (1) reading and showing the complex distribution of a sample (e.g., utility package); (2) performing numerical propagations of complex wavefields to provide in-focus DH and DHM images (e.g., numerical propagation package); (3) reconstructing the phase distribution of samples in in-line and slightly off-axis DH and DHM systems using PS techniques (e.g., phase-shifting package); and (4) reconstructing phase images in single-shot off-axis DHM systems operating in telecentric and non-telecentric configuration using automatic methods to estimate the best digital reference wavefront (e.g., fully-compensated phase reconstruction package). We have presented a sample code for each function implemented in the pyDHM library and validated its performance using simulated or experimental images. The pyDHM library is posted publicly on GitHub [65, 66]. The GitHub repository includes a complete documentation of the functions implemented, sample codes, and troubleshooting guidelines for correctly using the library. To increase the applicability of this library in our community, the GitHub repository also includes simulated and experimental holograms and some instructional videos on how to install and use the library [67–69]. In future works, we will expand the codes within our library and reduce its processing time using GPU implementations. Current implementations within the pyDHM library require that the users select the adequate reconstruction method based on the optical configuration of the DHM systems. Future work will focus on an automatic algorithm to reconstruct DHM holograms without prior knowledge of the DHM configuration (e.g., only hologram, source’s wavelength, and sensor’s pixel size). Because of the broad applicability of DHM systems in biology and medicine, we will create a graphical user interface (GUI) for the pyDHM library, aiming that users who lack coding skills and background in Optics and DHM could adopt this library. Such an app will allow the users to input a single hologram or a sequence/video of holograms, enabling the processing of the whole hologram series. Some reprocessing steps will be avoided, such as selecting the spatial filter mask to optimize such video processing. Our final goal is to expand such app further so that pyDHM can be used to analyze biological systems, including motility analysis for microorganism tracking and cell counting.