Services

Object detection is the process of detecting objects of interests in images or video frames using pre-trained networks. It is one of the main functionalities of a vision-based AI system. Recently, several innovations in the field of object detection has been introduced that provides robust and high performance solutions for the problem.

Object detection is between the most interesting services of fraime. With a simple interactive interface, one may try the two state-of-the-art networks, i.e. Detectron2 and YOLOv5, easily. For more technical discussion on the two networks, please refer to the object detection white paper.

Face detection is one of the main elements of all human-centered applications of artificial intelligence. Face is one of the most important biometrics in authentication systems. Face detection in real-world images is challenging due to different shapes, orientations, sizes, colors and occlusion due to wearing glasses, hat, and etc. In the last decade, a few promising algorithms have been proposed in this AI field of research.

In fraime Face Detection service, we provide two methods: 1. Dlib open source cross platform library, and 2. Multitask cascaded convolutional network (MTCNN) method. For more technical details of these methods, please refer to the corresponding white paper.

Face recognition refers to methods that are capable of matching an individual face from an image or video frame against a database of reference faces. As compared to other biometric techniques, facial recognition systems are able to perform mass identification and does not require the cooperation of the subject. Due to various orientations, scales and occlusion by glasses, masks and more, face recognition is between the most challenging problems in the field of biometric-based authentication. Correspondingly, a wide range of application-specific algorithms and tools have been provided in the literature and technology. For more technical discussion, please refer to face recognition white paper. At fraime, we provide a generic face recognition interface to examine this effective application.

Image inpainting is the process of filling regions of interest in images for generating a complete image from either a damaged, a deteriorating image or after removing an unwanted object. Recently deep neural networks have shown promising results for this challenging image processing task. Here, in fraime we propose a two-staged generative deep Image inpainting method called DeepFill v2. This method is capable of filling large and multiple areas of image without usual boundary artifacts, distorted structures and blurry textures inconsistent with surrounding areas that we observe in other deep networks. If you are interested for more technical information, please read the corresponding white paper.

Pose estimation refers to detecting keypoint locations that describe the overall shape (skeleton) of an object. In fraime, we provide human pose estimation to jointly detect human body, hand, facial, and foot keypoints in a single image. Two popular pose estimator are provided: OpenPose and MediaPipe. OpenPose is a multi-person bottom-up method while MediaPipe is top-down and in its standard version, it is single person. Different set of keypoints is extracted using each method that you may find the details in the corresponding white paper. Human pose estimation have wide applications such as motion analysis, action recognition and more.

Image annotation refers to the process of annotating images from a dataset with labels to train a machine learning model. Image annotation is a way to transfer human high-level knowledge of the image content to the model. Different models require different labeling. There are three common annotation types in machine vision applications: 1. Classification, 2. Object detection and 3. Image segmentation. fraime provides three annotation services to support these three annotation types: classification annotation, bounding box annotation and boundary annotation. For more technical information about the three types of annotations, please read this white paper. 

Deep learning is a sub-category of machine learning methods based on artificial neural networks that automatically extracts high-level features from the raw input data. Between different types of deep learning methods, convolutional neural networks (CNNs) are among the most successful and commonly used architectures in the deep learning applications. Therefore, in fraime two CNNs are provided for training on user specific dataset (i.e. Detectron2 and Yolov5). Please check the application of Detectron2 and YOLOv5 in Object Detection and Count Objects services. For more information about deep learning concepts, please refer to the corresponding white paper.

Anomaly detection addresses the challenging problem of detecting automatically exceptions or defects in a background image, i.e. identifying rare cases that differ from the typical cases that make up the majority of a dataset. In fraime, we investigated several types of Surface Defects and will add more and more anomalies in future. Multiple classification and segmentation deep networks are employed for each case and dataset. Please note that the training process was supervised now and the service has to be tested by images and defects similar to the ones in the dataset. Currently, 15 different datasets and their corresponding surface defect detection algorithm is ready for try. These datasets cover defect on surfaces like: Metal, Steel, Polymer and Texture. You may try the algorithms freely by our simple UI on your own images and in case of interest, you may leave us a customization request to retrain the networks on your specific dataset or other type of surfaces and defects.

Resolution enhancement refers to a group of techniques that are used in image processing or super-resolution microscopy for scaling up and improving low-resolution input images. Resolution enhancement has various applications, such as security and surveillance imaging, medical imaging, image generation, and satellite and astronomical imaging. There are both single and multiple image variants of resolution enhancement methods. fraime provides two single image resolution enhancement methods using deep learning: Residual Dense Network (RDN) and Residual in Residual Dense Network (RRDN). They both use residual learning that has also been widely adopted to ease the training process, either in image-level or feature-level. For more details of the two methods and network architecture, please refer to the white paper.

Count objects refers to automatically counting the number of interesting objects in images. It has several applications in industries, agriculture, medicine, etc. In fraime Count Object service two deep learning networks are provided: Detectron2 and YOLOv5. By default the two networks are trained over COCO dataset which consists 12 classes and 80 objects like humans, vehicles, animals, etc. For more technical information read Count Objects service white paper.

Line segment detection refers to detecting line segment primitives in images for various image analysis applications. Detecting image primitives may simplifies and accelerates many image processing applications like object detection, tracking, image segmentation, and etc. fraime Line Segment Detector employ the well-known LSD algorithm that following a statistical approach gives accurate results, a controlled number of false detections, and requires no parameter tuning. For more information about LSD technical details, please refer to Line Segment Detection white paper.

Arc (elliptical and circular arcs) and Line Segment Detection (ALD) refers to jointly detecting line segment, elliptical arc and circular arc primitives in images for various image analysis applications. Detecting image structures may simplify many image processing applications like object detection, tracking, image segmentation, etc. In fraime ArcLine Detector service, using a statistical algorithm which is called ELSDc, a fast and reliable algorithm, circular arcs, elliptical arcs and line segments are extracted from input image(s). For more information about ELSDc algorithm and its technical details, please refer to ALD white paper.

Contrast enhancement of poorly illuminated images have been widely used in many image processing applications where the subjective quality of images is important for human or machine interpretation. Contrast is the difference in visual properties that makes an object distinguishable from other objects and the background which is determined by the difference in the color and brightness. There are mainly two categories of image contrast enhancement algorithms: the global and the local enhancement. In fraime Contrast Enhancement demo, you can investigate a fast and robust local enhancement method which is called LLCC on your own images online. For more information on technical details of this method, please refer to the corresponding white paper.

Image denoising refers to restoring a clean image by removing undesirable noise distortions from input image. Image denoising is a classic image processing and computer vision problem and has been studied for decades while still is challenging. Numerous image denoiser have been provided from classic algorithms with mathematical principle to recent methods based on machine learning and deep learning advances. Using fraime interface you may try both approaches: a classic method which is called “Non-blind DCT-based method” and a deep learning method which is called “multi-stage learning”. For technical details of both algorithms please refer to the corresponding white paper.

Image deblurring refers to recovering a sharp image from the input blurred one. It is inherently an ill-posed problem that has been studied in many researches during the past few decades. The blurring of an image can be caused by many factors, like: camera or subject movement during the image acquisition, out-of-focus optics, or scattered light distortion. Numerous image deblurring methods have been provided from classic algorithms with mathematical principle to recent methods based on machine learning and deep learning advances. Using fraime interface you may try both approaches: a classic method which is called “Prior-based blind deblurring” and a deep learning method which is called “multi-stage learning”. For technical details of both algorithms please refer to the corresponding white paper.

Image deraining refers to removing undesirable raining artifacts from input images. Deraining as a pre-processing step may improve the performance of applications processing real-world images with raining artifacts. fraime provides two deep learning based deraining algorithms: multi-stage progressive image restoration network (MPRNet) and density-aware image deraining method using a multistream dense network (DID-MDN). For more technical details of the two networks, please refer to fraime deraining white paper.