Abstract: A method for image compression and a circuit system thereof are provided. In the method, pixel values of an image are obtained. A compression scenario is decided, for example, a uniform-quantization manner or a non-uniform-quantization manner is used for an M-bit image being compressed to an N-bit image so as to decide codeword sections for the image. Every codeword section has a codeword distance. The codeword sections have a fixed codeword distance in the uniform-quantization manner. Alternatively, in the non-uniform-quantization manner, the image can be divided into multiple codeword sections having different codeword distances according to a brightness distribution. Afterwards, a random number is generated for deciding codeword and index for original value of each of the pixels. An index table is accordingly formed. The index table is provided for obtaining the codeword in a decoding process by querying a codebook with the index.

Abstract: An image data augmentation device includes a memory and a processor. The memory is configured to store multiple commands and multiple images; the processor is connected to the memory to read the multiple images and execute multiple commands to: identify at least one object-related image including at least one object from the multiple images; capture at least one object image from the at least one object related image; and superimpose the at least one object image on multiple arbitrary positions in the multiple images to generate multiple training sample images, and using the multiple training sample images to execute machine learning. In addition, an image data augmentation method is also disclosed here.

Abstract: A method for processing image using fully connected convolutional neural network and a circuit system are provided. The method is operated using fully connected convolutional neural network (CNN) and performed by the circuit system. In the method, an image with a length, a width and an aspect ratio is obtained. A reference image closest to the input image can be obtained by querying a lookup table that records multiple reference images with various sizes to be adapted to the fully connected CNN. The input image can be resized as the closest reference image. A convolution operation is then performed onto the resized image, and a feature cube is formed after multiple operations of convolution. The feature cube is transformed to one-dimensional feature values that are configured to be inputted to a fully connected layer for fully connected operation. An output value of the fully connected CNN is generated.

Abstract: Disclosed is an image processing method based on sensor characteristics. The method includes the following steps: obtaining a constant of a relation “c??(x)/?{square root over (x)}” between output signals “x” and the noise standard deviation “?(x)” of the output signals, in which the output signals are outputted by a sensor based on a sensor gain such as an ISO value of a camera; and calculating an output value of a target pixel according to pixel values within a sampling window, the constant, and a front-end gain such as an auto exposure gain, in which the pixel values include an input value of the target pixel.

Abstract: A calculation method using pixel-channel shuffle convolutional neural network is provided. In the method, an operating system receives original input data. The original input data is pre-processed by a pixel shuffle process to be separated into multiple groups in order to minimize dimension of the data. The multiple groups of data are then processed by a channel shuffle process so as to form multiple groups of new input data selected for convolution operation. The unselected data are abandoned. Therefore, the dimension of the input data can be much effectively minimized. A multiplier-accumulator of the operating system is used to execute convolution operation using a convolution kernel and the multiple new groups of input data. Multiple output data are then produced.

Abstract: A motion image integration method includes acquiring a raw image, detecting a first motion region image and a second motion region image by using a motion detector according to the raw image, merging the first motion region image with the second motion region image for generating a motion object image according to a relative position between the first motion region image and the second motion region image, and cropping the raw image to generate a sub-image corresponding to the motion object image according to the motion object image. A range of the motion object image is greater than or equal to a total range of the first motion region image and the second motion region image. Shapes of the first motion region image, the second motion region image, and the motion object image are polygonal shapes.

Abstract: A motion detection method includes acquiring a raw image, detecting a motion object image according to the raw image by using a motion detector, cropping the raw image to generate a sub-image according to the motion object image, and inputting the sub-image to a processor for determining if a motion object of the sub-image matches with a detection category. The processor includes a neural network. The shape of the sub-image is a polygonal shape.

Abstract: A method and a system for pixel channel imbalance compensation for an image sensing circuit are provided. The system includes an image acquisition circuit having a lens, a color filter and an image sensor and a processing circuit. In the method performed by the processing circuit, a second frame image is retrieved from a motion image, and a first frame image that has undergone noise reduction can be retrieved from a memory. Motion detection is performed between the frames by comparing the first frame image and the second frame image. The motion detection is referred to as a reference for determining how to perform 3D noise reduction. A compensation value for channel imbalance between the adjacent channels can be estimated based on the image under noise reduction in a same buffer. While the pixel channel imbalance is compensated, the image is then restored by an interpolation method.

Abstract: An image processing circuit includes a receiving circuit, a feature fetching module and a decision circuit. In the operations of the image processing circuit, the receiving circuit is configured to receive image data. The feature fetching module is configured to use a multi-topological-convolutional network to fetch the features of the image data, to generate a plurality of image features determined by the characteristics and weights of the convolution filter, where the image features may be smooth features or edge features. In the present invention, the convolution filters used by the feature fetching module are not limited by a square convention filter, and the convolution filters may include the multiple topological convolutional network having non-square convolution filters. By using the multiple topological convolutional network of the present invention, the feature fetching module can fetch the rich image features for identifying the contents of the image data.

Abstract: Disclosed are a color reconstruction device and method capable of accurately recovering the missing color of a target pixel. The device includes: a direction-characteristic estimation circuit calculating a horizontal-variation characteristic and a vertical-variation characteristic according to a first color of a target pixel and the values of pixels within a reference range, in which the target pixel is in the reference range and a current value of the target pixel is a first color value; an edge-texture decision circuit determining which of N predetermined relations matches the relation between the horizontal-variation characteristic and the vertical-variation characteristic and thereby determining the directional characteristic of the target pixel, in which the N is not less than four; and a color recovery circuit calculating a second and a third color values of the target pixel according to the directional characteristic and the values of the pixels within the reference range.

Abstract: A pixel value calibration method includes: obtaining input image data generated by pixels, the input image data including a first group of pixel values in a first color plane and a second group of pixel values in a second color plane, generated by a first portion and a second portion of the pixels respectively; determining a difference function associated with filter response values and target values, the filter response values being generated by utilizing characteristic filter coefficients to filter first and second estimated pixel values of estimated pixel data in the first and second color planes, respectively; determining a set of calibration filter coefficients by calculating a solution of the estimated pixel data, the solution resulting in a minimum value of the difference function; and filtering the input image data, by a filter circuit using the set of calibration filter coefficients, to calibrate the first group of pixel values.

Abstract: A denoising method based on a signal-to-noise ratio (SNR), which includes: obtaining a current input coefficient; obtaining a current noise standard deviation by querying a first relationship table; querying a second relationship table according to the current noise standard deviation and the current input coefficient to obtain a current slope corresponding to the current input coefficient; generating a current output coefficient by multiplying the current input coefficient and a compression magnification function; and calculating the current output coefficient by substituting the current noise standard deviation, the current input coefficient, and the current slope.

Abstract: A method and a system for image correction are provided. The system includes an image processor that performs the method for image correction when receiving an input image. In the method, an overexposed area within the image is determined. The channel values of pixels in the overexposed area are applied with corresponding weights. An automatic white balance process is performed on the channel values of pixels in the overexposed area according to an expectation of low saturation in the overexposed area. Besides a first channel value of each of the pixels, a second channel value and a third channel value of the pixel can be estimated according to the channel values of periphery pixels. A correction value overexposed of the first channel value of the current pixel can be calculated according to the corresponding weights of the pixels in the overexposed area.

Abstract: An object detection method and an associated electronic device are provided, wherein the object detection method includes: utilizing an image processing circuit to determine whether motion occurs in an image to generate a determination result; selectively utilizing a specific bounding box to identify a target object to generate an identification result according to the determination result, wherein the specific bounding box represents a location of the target object in a previous image; and selectively updating information of the specific bounding box according to the identification result.

Abstract: Disclosed is an image processing method based on sensor characteristics. The method includes the following steps: obtaining a constant of a relation “c??(x)/?{square root over (x)}” between output signals “x” and the noise standard deviation “?(x)” of the output signals, in which the output signals are outputted by a sensor based on a sensor gain such as an ISO value of a camera; and calculating an output value of a target pixel according to pixel values within a sampling window, the constant, and a front-end gain such as an auto exposure gain, in which the pixel values include an input value of the target pixel.

Abstract: A method of updating background model for an image processing device is disclosed. The method includes receiving at least a background image, counting a number of occurrences of a color of each pixel of the background image according to a color information of each pixel of the at least a background image, to establish a background model including a statistical color information about the color and the number of occurrences of the color, receiving a current image, and determining whether to update the number of occurrences of the color in the background model corresponding to a color of each pixel of the current image according to a color information of each pixel of the current image and a result of comparing a random number with a threshold.

Abstract: A motion image integration method includes acquiring a raw image, detecting a first motion region image and a second motion region image by using a motion detector according to the raw image, merging the first motion region image with the second motion region image for generating a motion object image according to a relative position between the first motion region image and the second motion region image, and cropping the raw image to generate a sub-image corresponding to the motion object image according to the motion object image. A range of the motion object image is greater than or equal to a total range of the first motion region image and the second motion region image. Shapes of the first motion region image, the second motion region image, and the motion object image are polygonal shapes.

Abstract: A motion detection method includes acquiring a raw image, detecting a motion object image according to the raw image by using a motion detector, cropping the raw image to generate a sub-image according to the motion object image, and inputting the sub-image to a processor for determining if a motion object of the sub-image matches with a detection category. The processor includes a neural network. The shape of the sub-image is a polygonal shape.

Abstract: A method for processing image using fully connected convolutional neural network and a circuit system are provided. The method is operated using fully connected convolutional neural network (CNN) and performed by the circuit system. In the method, an image with a length, a width and an aspect ratio is obtained. A reference image closest to the input image can be obtained by querying a lookup table that records multiple reference images with various sizes to be adapted to the fully connected CNN. The input image can be resized as the closest reference image. A convolution operation is then performed onto the resized image, and a feature cube is formed after multiple operations of convolution. The feature cube is transformed to one-dimensional feature values that are configured to be inputted to a fully connected layer for fully connected operation. An output value of the fully connected CNN is generated.

Abstract: A method and a system for pixel channel imbalance compensation for an image sensing circuit are provided. The system includes an image acquisition circuit having a lens, a color filter and an image sensor and a processing circuit. In the method performed by the processing circuit, a second frame image is retrieved from a motion image, and a first frame image that has undergone noise reduction can be retrieved from a memory. Motion detection is performed between the frames by comparing the first frame image and the second frame image. The motion detection is referred to as a reference for determining how to perform 3D noise reduction. A compensation value for channel imbalance between the adjacent channels can be estimated based on the image under noise reduction in a same buffer. While the pixel channel imbalance is compensated, the image is then restored by an interpolation method.