An artificial intelligence based object identifying apparatus and a method thereof which are capable of easily identifying a type of an object in an image using a small size learning model are disclosed. According to an embodiment of the present disclosure, an object identifying apparatus configured to identify an object from an image includes a receiver configured to receive the image, an image modifier configured to modify the received image by predetermined methods to generate a plurality of modified images, and an object determinator configured to apply the plurality of modified images to a neural network trained to identify an object from the image to obtain a plurality of identification results and determine a type of an object in the received image based on the plurality of identification results.

A method for estimating text heights of text line images includes estimating a text height with a sequence recognizer. The method further includes normalizing a vertical dimension and/or position of text within a text line image based on the text height. The method may also further include calculating a feature of the text line image. In some examples, the sequence recognizer estimates the text height with a machine learning model.

The present disclosure provides a knowledge distillation based semantic image segmentation method including inputting an input image to a teacher network and a student network; normalizing a first feature vector corresponding to each pixel in a feature map of a last layer of the teacher network and normalizing a second feature vector corresponding to each pixel in a feature map of a last layer of the student network; generating the first channel and space association matrix and the second channel and space association matrix based on the normalized first feature vector and the normalized second feature vector, and defining a first loss function based on an Euclidean norm value of the difference between the first channel and space association matrix and the second channel and space association matrix.

A method for comparing videos of a surgical procedure is disclosed. The method comprising selecting a plurality of videos from a surgical video database. Each of the plurality of videos including video data of a first surgical procedure comprising a plurality of surgical steps. The method further including identifying a first surgical step included in the plurality of surgical steps within a first video segment in each of the plurality of videos. The method also including warping the first video segment to standardize a dimension of the first video segment in each of the plurality of videos.

An image recognition processing method and apparatus. The method comprises: obtaining original image data, convolutional neural network configuration parameters, and convolutional neural network operation parameters, the original image data comprising data of M pixel points (101); performing convolutional neural network operation on the original image data by a convolutional neural network operation module (2) according to the convolutional neural network configuration parameters and the convolutional neural network operation parameters, wherein the convolutional neural network operation module (2) comprises N operation components (21) which are provided in parallel, each operation component (21) comprises a convolution operation unit (22), a batch processing operation unit (23), and an activation operation unit (24) which are sequentially connected, the N operation components (21) respectively simultaneously perform convolution operation, batch processing operation, and activation operation on data of N pixel points in the original image data, and N is a positive integer smaller than or equal to M (102). The method improves the real-time performance of image recognition processing.

The invention relates to a method of reconstructing a fingerprint image from a plurality of fingerprint subimages captured by an optical microlens array fingerprint sensor, and a sensor system performing the method. In an aspect, a method of reconstructing a fingerprint image from a plurality of fingerprint subimages captured by an optical microlens array fingerprint sensor arranged under a touch-display of a device is provided. The method comprises capturing an image of a fingerprint with the fingerprint sensor, extracting, from the captured fingerprint image, a fingerprint subimage for each microlens in the array, normalizing the corrected fingerprint subimages with corresponding subimages of a uniformity calibration image with known uniformity properties, rotating the normalized fingerprint subimages 180 degrees while maintaining their relative position with respect to the captured fingerprint image from which the fingerprint subimages are extracted, and combining the rotated and normalized fingerprint subimages into a fingerprint image.

A system and a method for unveiling poorly visible or lightly colored nuclei in an input image are disclosed. An input image is fed to a color deconvolution module for deconvolution into two color channels that are processed separately before being combined. The input image is deconvolved into two separate images: a stain image and a counter stain image. A complement of the stain image is generated in order to clearly reflect the locations of the poorly visible or light-colored nuclei. The complement image and the counter stain image are optionally normalized and then combined and segmented, to generate an output image with clearly defined nuclei. Alternatively, the complement of the stain image and the counter stain image can optionally be normalized, and then segmented prior to being combined to generate the output image.

An image processing device 10 includes: a feature extraction unit 11 which obtains features in each of scaled samples of the region of interest in a probe image; a saliency generation unit 12 which computes the probabilities of the pixels in the scaled samples that contribute to the score or the label of the object of interest in the region; a dropout processing unit 13 which removes the features from the scaled samples which are not essential for the computing the score or the label of the object, using the computed probabilities.

Systems and methods for using a mobile device to submit an application for an insurance policy using images of documents captured by the mobile device are provided herein. The information is then used by an insurance company to generate a quote which is then displayed to the user on the mobile device. A user captures images of one or more documents containing information needed to complete an insurance application, after which the information on the documents is extracted and sent to the insurance company where a quote for the insurance policy can be developed. The quote can then be transmitted back to the user. Applications on the mobile device are configured to capture images of the documents needed for an insurance application, such as a driver's license, insurance information card or a vehicle identification number (VIN). The images are then processed to extract the information needed for the insurance application.

3D objects in the form of human faces are recognised at or adjacent an entrance to a building. Cameras are spaced horizontally from one another in an array to provide multiple viewpoints of a common scene with overlapping fields of view. For each of the cameras, a sequence of images of a face is generated and each of those images is normalised to a canonical view. One or more best normalised image for each camera is selected and feature data is extracted from the selected image. A recognition processor compares the extracted feature data of each selected image with stored, corresponding feature data of known 3D objects. A recognition result is output when the extracted feature data of at least one of the selected images corresponds to stored, corresponding feature data of at least one of the known 3D objects.