Quantization Process Research Articles

Nonlinear Quantization Method of SAR Images with SNR Enhancement and Segmentation Strategy Guidance

The quantization process of synthetic aperture radar (SAR) images faces significant challenges due to their high dynamic range, resulting in notable quantization distortion. This not only degrades the visual quality of the quantized images but also severely impacts the accuracy of image interpretation. To mitigate the distortion caused by uniform quantization and enhance visual quality, this paper introduced a novel nonlinear quantization framework via signal-to-noise ratio (SNR) enhancement and segmentation strategy guidance. This framework introduces guiding information to improve quantization performance in weak scattering regions. A histogram adjustment method is developed to incorporate the spatial information of SAR images into the quantization process to enhance the quantization performance, specifically within weak scattering regions. Additionally, the optimal quantizer is improved by refining the SNR distribution across quantization units, addressing imbalances in their allocation. Experimental results based on Gaofen-3 (GF-3) satellite data demonstrate that the proposed algorithm approaches the global quantization performance of optimal quantizers while achieving superior local quantization performance compared to existing methods.

Quantization via Distillation and Contrastive Learning.

Quantization is a critical technique employed across various research fields for compressing deep neural networks (DNNs) to facilitate deployment within resource-limited environments. This process necessitates a delicate balance between model size and performance. In this work, we explore knowledge distillation (KD) as a promising approach for improving quantization performance by transferring knowledge from high-precision networks to low-precision counterparts. We specifically investigate feature-level information loss during distillation and emphasize the importance of feature-level network quantization perception. We propose a novel quantization method that combines feature-level distillation and contrastive learning to extract and preserve more valuable information during the quantization process. Furthermore, we utilize the hyperbolic tangent function to estimate gradients with respect to the rounding function, which smoothens the training procedure. Our extensive experimental results demonstrate that the proposed approach achieves competitive model performance with the quantized network compared to its full-precision counterpart, thus validating its efficacy and potential for real-world applications.

Lossy Compression of Integer Astronomical Images Preserving Photometric Properties* *This work was supported in part by the Spanish Ministry of Science and Innovation (MICINN) and by the European Regional Development Fund (FEDER), funded by MCIN/AEI/10.13039/501100011033/FEDER, UE, under grant PID2021-125258OB-I00; also by the Catalan Government under grant SGR2021-00643; and also by the “PhotSat” project under the Institute for Space Studies of

Observatories are producing astronomical image data at quickly increasing rates. As a result, the efficiency of the compression methods employed is critical to meet the storage and distribution requirements of both observatories and scientists. This paper presents a novel lossy compression technique that is able to preserve the results of photometry analysis with high fidelity while improving upon the state of the art in terms of compression performance. The proposed compression pipeline combines a flexible bi-region quantization scheme with the lossless, dictionary-based, LPAQ9M encoder. The quantization process allows compression performance and photometric fidelity to be precisely tailored to different scientific requirements. A representative data set of 16-bit integer astronomical images produced by telescopes from all around the world has been employed to empirically assess its compression-fidelity trade-offs, and compare them to those of the de facto standard Fpack compressor. In these experiments, the widespread SExtractor software is employed as the ground truth for photometric analysis. Results indicate that after lossy compression with our proposed method, the decompressed data allows consistent detection of over 99% of all astronomical objects for all tested telescopes, maintaining the highest photometric fidelity (as compared to state of the art lossy techniques). When compared to the best configuration of Fpack (Hcompress lossy using 1 quantization parameter) at similar compression rates, our proposed method provides better photometry precision: 7.15% more objects are detected with magnitude errors below 0.01, and 9.13% more objects with magnitudes below SExtractor’s estimated measurement error. Compared to the best lossless compression results, the proposed pipeline allows us to reduce the compressed data set volume by up to 38.75% and 27.94% while maintaining 90% and 95%, respectively, of the detected objects with magnitude differences lower than 0.01 mag; and up to 18.93% while maintaining 90% of the detected objects with magnitude differences lower than the photometric measure error.

Study of Weight Quantization Associations over a Weight Range for Application in Memristor Devices

The development of hardware-based cognitive computing systems critically hinges upon the integration of memristor devices capable of versatile weight expression across a spectrum of resistance levels while preserving consistent electrical properties. This investigation aims to explore the practical implementation of a digit recognition system utilizing memristor devices with minimized weighting levels. Through the process of weight quantization for digits represented by 25 or 49 input signals, the study endeavors to ascertain the feasibility of digit recognition via neural network computation. The integration of memristor devices into the system architecture is poised to streamline the representation of the resistors required for weight expression, thereby facilitating the realization of neural-network-based cognitive systems. To minimize the information corruption in the system caused by weight quantization, we introduce the concept of “weight range” in this work. The weight range is the range between the maximum and minimum values of the weights in the neural network. We found that this has a direct impact on weight quantization, which reduces the number of digits represented by a weight below a certain level. This was found to help maintain the information integrity of the entire system despite the reduction in weight levels. Moreover, to validate the efficacy of the proposed methodology, quantized weights are systematically applied to an array of double-layer neural networks. This validation process involves the construction of cross-point array circuits with dimensions of 25 × 10 and 10 × 10, followed by a meticulous examination of the resultant changes in the recognition rate of randomly generated numbers through device simulations. Such endeavors contribute to advancing the understanding and practical implementation of hardware-based cognitive computing systems leveraging memristor devices and weight quantization techniques.

Construction Jobsite Image Classification Using an Edge Computing Framework

Image classification is increasingly being utilized on construction sites to automate project monitoring, driven by advancements in reality-capture technologies and artificial intelligence (AI). Deploying real-time applications remains a challenge due to the limited computing resources available on-site, particularly on remote construction sites that have limited telecommunication support or access due to high signal attenuation within a structure. To address this issue, this research proposes an efficient edge-computing-enabled image classification framework for support of real-time construction AI applications. A lightweight binary image classifier was developed using MobileNet transfer learning, followed by a quantization process to reduce model size while maintaining accuracy. A complete edge computing hardware module, including components like Raspberry Pi, Edge TPU, and battery, was assembled, and a multimodal software module (incorporating visual, textual, and audio data) was integrated into the edge computing environment to enable an intelligent image classification system. Two practical case studies involving material classification and safety detection were deployed to demonstrate the effectiveness of the proposed framework. The results demonstrated the developed prototype successfully synchronized multimodal mechanisms and achieved zero latency in differentiating materials and identifying hazardous nails without any internet connectivity. Construction managers can leverage the developed prototype to facilitate centralized management efforts without compromising accuracy or extra investment in computing resources. This research paves the way for edge “intelligence” to be enabled for future construction job sites and promote real-time human-technology interactions without the need for high-speed internet.

Transmit Precoding via Block Diagonalization with Approximately Optimized Distance Measures for Limited Feedback in Dense Cellular Networks with Multiantenna Base Stations

This study introduces distance metrics for quantized-channel-based precoding in multiuser multiantenna systems, aiming to enhance spectral efficiency in dense cellular networks. Traditional metrics, such as the chordal distance, face limitations when dealing with scenarios involving limited feedback and multiple receive antennas. We address these challenges by developing distance measures that more accurately reflect network conditions, including the impact of intercell interference. Our distance measures are specifically designed to approximate the instantaneous rate of each user by estimating the unknown components during the quantization stage. This approach enables the associated users to efficiently estimate their achievable rates during the quantization process. Our distance measures are specifically designed for block diagonalization precoding, a method known for its computational efficiency and strong performance in multi-user multiple-input and multiple-output systems. The proposed metrics outperform conventional distance measures, particularly in environments where feedback resources are constrained, as is often the case in 5G and emerging 6G networks. The enhancements are especially significant in dense cellular networks, where accurate channel state information is critical for maintaining high spectral efficiency. Our findings suggest that these new distance measures offer a robust solution for improving the performance of limited-feedback-based precoding in cellular networks.

Joint contrastive self-supervised learning and weak-orthogonal product quantization for fast image retrieval

Hashing and quantization methods based on deep learning are being widely used in large-scale image retrieval. Most methods use traditional supervised models to train models. Although these methods achieve excellent performance, they rely too much on expensive label information and do not fully utilize data resources. These methods overly encourage binary codes to retain original information, which may lead to models building too much useless information at the expense of discriminative information that is more important for retrieval. To address these issues, we propose a self-supervised method that does not require labeled information, called Contrastive Self-Supervised Weak-Orthogonal Product Quantization (CSWPQ). The method is trained in a label-free self-supervised manner to reduce the need for labels by maximizing the similarity of different views of the same image and minimizing the similarity of different image views, which not only improves the generalization ability of the model but also enhances the robustness. We introduce a quantified contrastive learning method that is different from traditional contrastive learning to learn more discriminative image representations by constructing quantified image features of different views of the image. In addition, in order to reduce the quantization error caused by product quantization, we impose a weak-orthogonal constraint to increase the diversity of representations, reduce the information loss and computational overhead in the quantization process, and improve the computational speed. Extensive experiments on CIFAR-10, NUS-WIDE and FLICKR25K datasets show that our proposed method achieves better performance.

Lossy Image Compression with Stochastic Quantization

Introduction. Lossy image compression algorithms play a crucial role in various domains, including graphics, and image processing. As image information density increases, so do the resources required for processing and transmission. One of the most prominent approaches to address this challenge is color quantization, proposed by Orchard et al. (1991). This technique optimally maps each pixel of an image to a color from a limited palette, maintaining image resolution while significantly reducing information content. Color quantization can be interpreted as a clustering problem (Krishna et al. (1997), Wan (2019)), where image pixels are represented in a three-dimensional space, with each axis corresponding to the intensity of an RGB channel. The purpose of the paper. Scaling of traditional algorithms like K-Means can be challenging for large data, such as modern images with millions of colors. This paper reframes color quantization as a three-dimensional stochastic transportation problem between the set of image pixels and an optimal color palette, where the number of colors is a predefined hyperparameter. We employ Stochastic Quantization (SQ) with a seeding technique proposed by Arthur et al. (2007) to enhance the scalability of color quantization. This method introduces a probabilistic element to the quantization process, potentially improving efficiency and adaptability to diverse image characteristics. Results. To demonstrate the efficiency of our approach, we present experimental results using images from the ImageNet dataset. These experiments illustrate the performance of our Stochastic Quantization method in terms of compression quality, computational efficiency, and scalability compared to traditional color quantization techniques. Conclusions. This study introduces a scalable algorithm for solving the color quantization problem without memory constraints, demonstrating its efficiency on a subset of images from the ImageNet dataset. The convergence speed of the algorithm can be further enhanced by modifying the update rule with alternative methods to Stochastic Gradient Descent (SGD) that incorporate adaptive learning rates. Moreover, the stochastic nature of the proposed solution enables the utilization of parallelization techniques to simultaneously update the positions of multiple quants, potentially leading to significant performance improvements. This aspect of parallelization and its impact on algorithm efficiency presents a topic for future research. The proposed method not only addresses the limitations of existing color quantization techniques but also opens up new possibilities for optimizing image compression algorithms in resource-constrained environments. Keywords: non-convex optimization, stochastic optimization, stochastic quantization, color quantization, lossy compression.

Fast view-specific generation of high-definition holograms with enhanced quality.

Holography is often considered as the most promising immersive technology because it provides all the depth cues of the human visual system. Some limitations still need to be overcome such as the huge computational load of high-definition holograms and the noise introduced in the reconstructed scene during the quantization process. In this paper, we propose what we believe is a novel view-specific layer-based stereogram approach combined with a view-dependent error diffusion algorithm that aims to solve those limitations. This method selects the light waves of the 3D scene that reach a specific viewing area and leverages this particular configuration to apply an error diffusion algorithm. Two additional quality enhancement features are observed: the reduction of the conjugate order perceptibility and the increased brightness of the reconstructed scene. Numerical and optical experiments demonstrate the time savings and quality enhancements of our approach.

EfficientQ: An efficient and accurate post-training neural network quantization method for medical image segmentation

Model quantization is a promising technique that can simultaneously compress and accelerate a deep neural network by limiting its computation bit-width, which plays a crucial role in the fast-growing AI industry. Despite model quantization’s success in producing well-performing low-bit models, the quantization process itself can still be expensive, which may involve a long fine-tuning stage on a large, well-annotated training set. To make the quantization process more efficient in terms of both time and data requirements, this paper proposes a fast and accurate post-training quantization method, namely EfficientQ. We develop this new method with a layer-wise optimization strategy and leverage the powerful alternating direction method of multipliers (ADMM) algorithm to ensure fast convergence. Furthermore, a weight regularization scheme is incorporated to provide more guidance for the optimization of the discrete weights, and a self-adaptive attention mechanism is proposed to combat the class imbalance problem. Extensive comparison and ablation experiments are conducted on two publicly available medical image segmentation datasets, i.e., LiTS and BraTS2020, and the results demonstrate the superiority of the proposed method over various existing post-training quantization methods in terms of both accuracy and optimization speed. Remarkably, with EfficientQ, the quantization of a practical 3D UNet only requires less than 5 min on a single GPU and one data sample. The source code is available at https://github.com/rongzhao-zhang/EfficientQ.

Generalized Brans-Dicke cosmology, noether symmetries and quantization

A generalized Brans-Dicke theory of gravitation is applied to investigate a cosmological model for a primordial universe that includes a self-interacting bosonic field as source. This model permits, initially, to consider an arbitrary ω function, not usual for BD formulations. The imposition of Noether symmetries conditions identifies a reduced space for the ω function. This model is shown to present a variety of behaviors consistent with a primordial era, depending on the parameters chosen. We find that the bosonic field is capable of promoting a transition from an initially accelerated to a decelerated regime. The self-interaction potential is also restricted by the Noether conditions, and can be interpreted as a vacuum energy contribution. The model also permits a quantization process, going through a hamiltonian formulation and recognizing the presence of cyclic variables. The Bohm approach shows as result a consistent model with a smooth classical limit as the Universe leaves the Planck era.

Feature value quantization and reduction process for predicting heart attack possibility and the level of severity by a machine learning model

Heart disease is a prevalent condition nowadays that, if undiagnosed, can be deadly. To predict heart disease, researchers designed many machine learning models. In this study, we propose a model that chooses fewer attribute columns for training, and we use these chosen features to determine the heart problem severity. Correlation Repeated Heat map and Information Gain were used for selecting the features. To train our model we used the UCI Cleveland heart disease dataset. We removed duplicate data to improve the accuracy score, and we also encoded the categorical data collection using the OneHot(OH) encoding method, which can improve prediction accuracy. Support Vector, Logistic Regression, K-Nearest Neighbour, Naive Bayes, Decision Tree, Random Forest, Adaboost, and XGBoost are the eight classifier algorithms that are used in this process overall. Based on repeated heat map correlation, we compare the accuracy score each time. In this proposed method, the Adaboost classification algorithm used by the fbs row heat map achieves the highest accuracy for heart disease detection and it is 92%. By choosing features according to the information gain value, we compare the accuracy score each time in information gain. For both XGBoost and Logistic Regression, we got an accuracy score of 93.44%. However, compared to the XGBoost classification technique, Logistic Regression requires less time. Accuracy, precision, recall, f1-score, sensitivity, specificity, and the AUC of ROC charts were used to evaluate the performance of the model. Overall, the results of our model demonstrate that it is reliable and accurate in identifying cardiac disease and its level of severeness.

An Adversarial Example Method Based on Dropping Information to Improve Attack Capability

At present, deep neural network has been widely used in many research fields. However, with the in-depth research on artificial intelligence research, it is found that artificial intelligence technology based on deep neural networks bringsconvenience coming with potential security risks. For example, an attacker may misguide the image classification model to output a wrong result with high confidence by adding slight perturbations to a clean image via the adversarial example method. Compared with the previous methods of adding additional information to the images to generate adversarial examples, Ranjie Duan et al. proposed the AdvDrop algorithm. In this algorithm, adversarial examples are generated by deleting the existing information of the images, which is realized by adjusting the quantization step. However, in the quantization process, the AdvDrop algorithm does not consider the different effects of different gradient values of the quantization table on the adversarial effect. In this regard, AdvDrop+ is proposed, that is, in each iteration, the quantization tables are updated according to the gradient numerical values scaled with a factor. To find the proper scaling factor, we find the gradient value of the highest frequency in the gradient histogram and compute its logarithm and the final result is the scaling factor. The experiments show that AdvDrop+ has better attack performance than AdvDrop under the setting of target attack with nearly the same image distortion. At the same time, AdvDrop+ retains the characteristic of AdvDrop, which can drop information.

Adaptive Distributed Heterogeneous Formation Control for UAV-USVs with Input Quantization

This paper investigates the cooperative formation trajectory tracking problem for heterogeneous unmanned aerial vehicle (UAV) and multiple unmanned surface vessel (USV) systems with input quantization performance. Firstly, at the kinematic level, a distributed guidance law based on an extended state observer (ESO) is designed to compensate for the unknown speed of neighbor agents for expected trajectory tracking, and subsequently at the dynamic level, an ESO is utilized to estimate model uncertainties and environmental disturbances. Following that, a linear analytic model is employed to depict the input quantization process, and the corresponding adaptive quantization controller is designed without necessitating prior information on quantization parameters. Based on the input-to-state stability, the stability of the proposed control structure is proved, and all the signals in the closed-loop system are ultimately bounded. Finally, a simulation study is provided to show the efficacy of the proposed strategy.

A Neural-Network-Based Watermarking Method Approximating JPEG Quantization.

We propose a neural-network-based watermarking method that introduces the quantized activation function that approximates the quantization of JPEG compression. Many neural-network-based watermarking methods have been proposed. Conventional methods have acquired robustness against various attacks by introducing an attack simulation layer between the embedding network and the extraction network. The quantization process of JPEG compression is replaced by the noise addition process in the attack layer of conventional methods. In this paper, we propose a quantized activation function that can simulate the JPEG quantization standard as it is in order to improve the robustness against the JPEG compression. Our quantized activation function consists of several hyperbolic tangent functions and is applied as an activation function for neural networks. Our network was introduced in the attack layer of ReDMark proposed by Ahmadi et al. to compare it with their method. That is, the embedding and extraction networks had the same structure. We compared the usual JPEG compressed images and the images applying the quantized activation function. The results showed that a network with quantized activation functions can approximate JPEG compression with high accuracy. We also compared the bit error rate (BER) of estimated watermarks generated by our network with those generated by ReDMark. We found that our network was able to produce estimated watermarks with lower BERs than those of ReDMark. Therefore, our network outperformed the conventional method with respect to image quality and BER.

A 14-Bit Hybrid Analog-to-Digital Converter for Infrared Focal Plane Array Digital Readout Integrated Circuit.

This paper presents a 14-bit hybrid column-parallel compact analog-to-digital converter (ADC) for the application of digital infrared focal plane arrays (IRFPAs) with compromised power and speed performance. The proposed hybrid ADC works in two phases: in the first phase, a 7-bit successive approximation register (SAR) ADC performs coarse quantization; in the second phase, a 7-bit single-slope (SS) ADC performs fine quantization to complete the residue voltage conversion. In this work, the number of unit capacitors is reduced to 1/128th of that of a conventional 14-bit SAR ADC, which is beneficial for the application of small pixel-pitch IRFPAs. In this work, a tradeoff segmented thermometer-coded digital-to-analog converter (DAC) is adopted in the first 7-bit coarse quantization process: the lower 3-bit is binary coded, and the upper 4-bit is thermometer coded. A thermometer-coded DAC can improve the linearity of ADC. Capacitor array matching can be incredibly relaxed compared with a binary-weight 14-bit SAR ADC, resulting in a noncalibration feature. Moreover, by sharing DAC and comparator analog circuits between the SAR ADC and the SS ADC, the power consumption and layout area are consequently reduced. The proposed hybrid ADC was fabricated using a 180 nm CMOS process. The measurement results show that the proposed ADC has a differential nonlinearity of -0.61/+0.84 LSB and a sampling rate of 120 kS/s. The developed ADC achieves a temporal noise of 1.7 LSBrms at a temperature of 77 K. In addition, the SNDR is 72.9 dB, and the ENOB is 11.82 bit, respectively. Total power consumption is 71 μW from supply voltages of 3.3 V (analog) and 1.8 V (digital).

A lossless quantization approach for physical-layer key generation in vehicular ad hoc networks based on received signal strength

Vehicular Ad Hoc Networks (VANETs) provide various benefits and play a crucial role in improving efficiency and ensuring human safety across different applications. However, these advantages also give rise to security challenges and privacy concerns, necessitating a thorough examination of security attacks. Current key generation schemes, particularly those based on the Physical-Layer Model (PLM), face limitations such as low key generation rates and inadequate randomness. This paper introduces an innovative key generation method that utilizes adaptive physical-layer techniques and lossless quantization. The method involves an eight-level quantization process, which enables precise granularity and adaptive selection of quantization thresholds to tailor the computation of Received Signal Strength (RSS) measurements to individual needs. The adaptive approach ensures the retention of information within RSS measurements, resulting in reduced bit disagreement rates and enhanced randomness of the generated keys. Simulated evaluations demonstrate the effectiveness of the proposed method, showing superior performance in terms of bit generation, entropy, and secrecy rates, while also minimizing the occurrence of unnecessary measurements. This advancement holds significant promise for strengthening the security framework within VANETs.

A Convolutional Neural Network-Based Quantization Method for Block Compressed Sensing of Images.

Block compressed sensing (BCS) is a promising method for resource-constrained image/video coding applications. However, the quantization of BCS measurements has posed a challenge, leading to significant quantization errors and encoding redundancy. In this paper, we propose a quantization method for BCS measurements using convolutional neural networks (CNN). The quantization process maps measurements to quantized data that follow a uniform distribution based on the measurements' distribution, which aims to maximize the amount of information carried by the quantized data. The dequantization process restores the quantized data to data that conform to the measurements' distribution. The restored data are then modified by the correlation information of the measurements drawn from the quantized data, with the goal of minimizing the quantization errors. The proposed method uses CNNs to construct quantization and dequantization processes, and the networks are trained jointly. The distribution parameters of each block are used as side information, which is quantized with 1 bit by the same method. Extensive experiments on four public datasets showed that, compared with uniform quantization and entropy coding, the proposed method can improve the PSNR by an average of 0.48 dB without using entropy coding when the compression bit rate is 0.1 bpp.

МЕТОД КОДУВАННЯ ВІДЕОДАНИХ В СПЕКТРАЛЬНО- ПАРАМЕТРИЧНОМУ ПРОСТОРІ

It is shown that the development of modern infocommunication systems is followed by the trend of widespread use of wireless technologies. The level of intellectualization of the protocols of their functioning is increasing. Accordingly, the characteristics of wireless infocommunication technologies for data transmission are increasing. This contributes to their use for data transmission from remote robotic platforms in the process of information support for the functioning of critical infrastructure. It is substantiated that the transmission of video data with the required level of bit intensity on the basis of existing on-board telecommunication technologies is organized under conditions of significant time delays. Consequently, the level of efficiency of information delivery is lost. On the other hand, there are losses in the reliability of video information. It is argued that the option for solving the problem lies in the further development of technologies for encoding video information streams. The most common platform is the JPEG format. However, the city has disadvantages. They refer to the loss of compression levels in the event of a reduction in the potential for detecting psychovisual addictions. The article substantiates an approach for improvement, which consists in the use of an adapted coding technology in the spectral-parametric space. This approach consists in detecting subbands in spectral arrays and their subsequent parameterization with the formation of structural components at a significant level and length. Further processing is proposed to be carried out taking into account the informativeness of segments and reducing the amount of structural redundancy. An approach for constructing a technology for compression of spectral arrays with the detection of spectral sub-bands is substantiated. A model has been created to find the amount of information for transformed segments in the spectral-parametric space. This takes into account the available amount of redundancy, which is established by determining the constraints on the values of structural components. A method for encoding segments in spectral-parametric description has been developed. The main basic components of the method relate to the processing of sets of two-component structural-parametric components, taking into account the peculiarities of their content depending on the presence of quantization processes.

Local Means Binary Networks for Image Super-Resolution.

The success of modern single image super-resolution (SISR) algorithms is inspired by the development of deep convolutional neural networks (CNNs). However, these CNN-based methods require considerable computation and complexity, making it impossible for these methods to perform real-time calculations in edge devices. Thus, lightweight model design has become a development trend in the super-resolution field, including pruning, quantization, and other methods. The 1-bit quantization is an extreme lightweight method which can reduce the calculation amount of the model in an extreme manner and is friendly to hardware such as edge devices. Most existing binary quantization approaches lead to a large information loss during forward propagation, especially in detailed color information (e.g., edge, texture, and contrast). The loss of color information makes modern binary methods unsuitable for SISR tasks. We think the loss occurs because these methods typically utilize a uniform threshold to quantize the weights and activations. Thus, in this article, we thoroughly analyze the difference between normal classification tasks and SISR tasks, and present a binarization scheme based on local means. The proposed method can maintain more detailed information in feature maps using dynamic thresholds during quantization. Specifically, each value in the full precision activations has a corresponding threshold during the quantization process, and those thresholds are determined by the full precision values of the surroundings. In addition, a gradient approximator is introduced to adaptively optimize the gradient for updating binary weights. We then verify the effectiveness of our method for training binary networks on several SISR benchmarks including VDSR and SRResNet. Experimental results show that the proposed method can outperform the state-of-the-art algorithms to obtain binary networks for image super-resolution with better peak signal-to-noise ratio (PSNR) values and visual quality.