Siemens Aktiengesellschaft

Ringcentral, Inc.

Unify Gmbh Co. Kg

Siemens Enterprise Communications Gmbh & Co. Kg

Peter Amon

Andreas Hutter

Norbert Oertel

André Kaup

Bernhard Agthe

Jürgen Pandel

Gero Bäse

Robert Kutka

Uwe Rauschenbach

Benoit Timmermann

Sanjukta Ghosh

Andreas Weinlich

Eugen Wige

Thomas Rathgen

Michael Kapralov

Johannes Rehm

Marcus Laumer

Agthe Bernhard

A temporal sequence of pictures is generated in a method for encoding of a first video stream. To do so, a synchronization signal can be used, which can be derived from a second video stream independently of the first video stream. Alternatively, the encoding of a second video stream independent of the first video stream can be based on the same principle as for the encoding of the first video stream.

Method for encoding of a video stream

Methods for generating a deep neural net and for localizing an object in an input image, the deep neural net, a corresponding computer program product, and a corresponding computer-readable storage medium are provided. A discriminative counting model is trained to classify images according to a number of objects of a predetermined type depicted in each of the images, and a segmentation model is trained to segment images by classifying each pixel according to what image part the pixel belongs to. Parts and/or features of both models are combined to form the deep neural net. The deep neural net is adapted to generate, in a single forward pass, a map indicating locations of any objects for each input image.

Methods for generating a deep neural net and for localising an object in an input image, deep neural net, computer program product, and computer-readable storage medium

The present embodiments relate to analysing an image. An artificial deep neural net is pre-trained to classify images into a hierarchical system of multiple hierarchical classes. The pre-trained neural net is then adapted for one specific class, wherein the specific class is lower in the hierarchical system than an actual class of the image. The image is then processed by a forward pass through the adapted neural net to generate a processing result. An image processing algorithm is then used to analyse the processing result focused on features corresponding to the specific class.

Method and apparatus for analysing an image

The disclosure relates to a method and an apparatus for analyzing an image using a deep neural net pre-trained for multiple classes. The image is processed by a forward pass through an adapted neural net to generate a processing result. The adapted neural net is adapted from the pre-trained neural net to focus on exactly one selected class. The processing result is then analyzed focused on features corresponding to the selected class using an image processing algorithm. A modified image is generated by removing a manifestation of these features from the image.

Method and apparatus for analyzing an image

A method and apparatus for detecting objects of interest in images, the method comprising the steps of supplying (S) at least one input image to a trained deep neural network, DNN, which comprises a stack of layers; and using at least one deconvolved output of at least one learned filter or combining (S) deconvolved outputs of learned filters of at least one layer of the trained deep neural network, DNN, to detect the objects of interest in the supplied images.

Method and apparatus for detecting objects of interest in images

Functional data structure for coding a set of digital images. Each of the images is compressed into at least-one first data stream portion, which comprises a portion of the macroblocks reduced by physical redundancies, and-one second data stream portion, which describes the redundancies. For the intraprediction macroblocks, the first data stream portion can be reduced by color value statements with correlations to color values from rows of pixels outside and at one edge of the intraprediction macroblock and for which, in the case of pixels outside the compressed image, a color value default is assumed. The second data stream can comprise intrapredictors for describing the correlations, with coding of an area which is divided into first areas, each of which is occupied by the macroblocks of one of the digital images, and a second area which spaces apart the first areas and is occupied by default color value pixels.

Multiplex method and associated functional data structure for combining digital video signals

In a method for coding a sequence of digital images, a prediction error between predicted values and the original values of pixels is processed for generating the coded sequence of digital images. A preset prediction mode is an intra-prediction mode based on pixels of a single image and includes, for a region of pixels with reconstructed values in the single image and for a template of an image area, comparing a first patch of pixels in the region that surround a first pixel to be predicted based on the template with several second patches. A predicted value of the first pixel is determined based on the values of one or more second pixels that have the highest similarity described by the similarity measure among all second pixels of the plurality of second pixels in the region.

Method for coding sequence of digital images

An encoder encodes a video signal formed of video frames, each including image blocks. The encoder includes a processing unit which calculates at least one high resolution reference image block on the basis of previously encoded image blocks by executing a super resolution algorithm to perform a local motion compensation; and a motion compensation unit which calculates on the basis of the calculated high resolution reference image block a temporal predictor which is subtracted from a current image block of the video signal. Together, the encoder and a corresponding decoder improve the signal quality of a video signal significantly.

Method and apparatus for encoding and decoding video signal

For each array of pixels, an autoregressive pixel-prediction method is performed based on a weighted sum of reconstructed pixel values of reconstructed pixels in a specific neighborhood region adjacent to the current pixel to be coded. For determining the weights, the pixel values in a specific training region adjacent to the current pixel are taken into account. The coding method is characterized by an appropriate determination of the specific neighborhood region and the specific training region in case that reconstructed pixel values do not exist for all pixels in the neighborhood region and the training region. In such a case, the number of pixels in the neighborhood region is reduced to a number of reconstructed pixels until the ratio between the number of pixels in the training region and the number of pixels in the neighborhood region exceeds a predetermined threshold.

Autogressive pixel prediction in the neighborhood of image borders

In a method for coding a sequence of digitized images, a motion compensation process is performed, the motion compensation process using motion vectors between image blocks referring to a number of pixels in a current image and reference blocks referring to a number of pixels in a reference image. The reference image is based on one or more images out of the sequence. For each image block of at least a part of the current image a temporal prediction based on a corresponding reference block indicated by a motion vector is performed, resulting in a prediction error between the image block and the corresponding reference block, where the prediction error is coded. In the method of the invention, the current image in the sequence is divided into several image areas, with a reference area in the reference image being associated with each image area. The temporal prediction of an image block in an image area is based on a reference block at least partially located in the reference area associated with the image area and including pixel information from this reference area.

Method for coding a sequence of digitized images

Methods and devices transform image data, which are transformed by a compression filter before being compressed and stored in a reference image memory. In an extension, an inverse transformation to that of the compression filter is performed by a decompression filter when image data from the reference memory are read out and decompressed. The methods and devices can be used for image compression methods and image decompression methods that use reference image memories.

Compression and decompression of reference images in a video encoder

The embodiments relate to pixel-prediction methods and devices that are based on one of a selection of one out of at least two model functions that provide a prediction function and an adaptive model function that is capable to predict intensity characteristics of reference pixels used for prediction.

Pixel-prediction for compression of visual data

For a compressed video stream formed of a sequence of compressed data packets, each compressed packet is assigned to a frame by processing the undecoded data packet. A frame includes packets. A first plurality of frames is assigned to a first sliding window and a second plurality of frames is assigned to a second sliding window. The first and second sliding windows have different frames. A first window indicator is calculated by determining a first indicator representing an average number of packets per frame within the first window, and/or a second indicator representing size information of the data within the first window. A second window indicator is calculated by determining a third indicator representing an average number of packets per frame within the second window, and/or a fourth indicator representing size information of the data within the second window. The first and the second window indicators are compared.

Growing community of inventors

Munich, Germany

Peter Amon