JPH04305778A - Color data converter - Google Patents

Abstract

PURPOSE:To shorten time required for learning in a neural network having multilayer structure. CONSTITUTION:This neural network color data converter consists of an input layer 1, an intermediate layer 2 and a main layer 3. Each layer consists of many neurons, executes the input processing of data inputted from another device and outputs the processed data to the other layer. The strength of neuron connection between respective layers is called weight and indicates the weight of inputted data. In the neural network of the multilayer structure having a unit previously weighted so as to convert color data in the intermediate layer, frequency necessary for learning can be reduced by preparing the previously weighted unit and the learning time can be shortened.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a color data conversion apparatus using a multilayer neural network for color data conversion.

0002

2. Description of the Related Art Conventionally, in a multilayer neural network, the neural network must be trained in advance in order to perform data conversion.

0003

[Problems to be Solved by the Invention] However, in general, a backpropagation type neural network, which is a typical example of a multilayer neural network, requires an enormous number of training sessions. For this reason, in the past, converting color data using a neural network required an enormous number of learning cycles, which posed the problem of a long learning time.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a means for shortening the time required for learning in order to solve these conventional problems.

[0005]

[Means for Solving the Problems] In order to solve the above problems, the present invention has a configuration in which the middle layer of the neural network has means for setting weights in advance to convert color data, thereby improving learning. We have made it possible to significantly reduce the number of times.

[0006]

[Operation] A neural network in which weights are set in advance in the intermediate layer operates to perform color data conversion at the time learning is started. The only learning that is actually required is for final adjustments to fit the actual situation.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Examples of the present invention will be described below with reference to the drawings. FIG. 1 shows the configuration of a neural network color image conversion device according to the present invention. 1 is an input layer of the neural network, 2 is an intermediate layer of the neural network, and 3 is an output layer of the neural network. Each layer consists of multiple neurons, each receiving data from neurons in other layers. When receiving data, set the weight from which neuron to give more weight to the data received. The weight is
Set the value determined in advance.

[0008] To find the value of the weight to be set, proceed as follows. FIG. 2 is a configuration diagram of a neural network for determining the value of the weight to be set. Learning data 5 is input to the neural network 4 that calculates weight values. Learning data 5 is red, blue, and green components of one color. At this time, output data 6 of the neural network is obtained. This data is interpreted as data indicating brightness, hue, and vividness, and if the result is not the expected result, teacher data 7 is input. The teacher data 7 is data indicating the brightness, hue, and vividness of the input color. This allows the neural network to learn. This process is repeated until the correct brightness, hue, and vividness can be output, and learning for one color is completed. Do this study for different colors.

[0009] The target weights can be determined by the weights of the neural network that has completed learning for various colors. Once you have calculated this weight,
In the neural network according to the present invention, there is no need to re-learn, just set weights to the corresponding neurons.

0010

[Effects of the Invention] As explained above, this invention has a structure in which the weights of the network that have been trained in advance are set in the middle layer of the neural network, so that it is possible to obtain the expected output with a small number of training times. Neural networks can be trained. As a result, the time required for learning was significantly reduced.

[Brief explanation of the drawing]

FIG. 1 is a configuration diagram of a neural network color conversion device according to the present invention.

FIG. 2 is a configuration diagram of a neural network that calculates weight values to be set.

[Explanation of symbols]

1 Input layer of neural network 2 Middle layer of neural network 3 Output layer of neural network 4 Neural network

Claims (1)

[Claims] Claim 1: A neural network having a multilayer structure having an intermediate layer, comprising means for initializing weighting so that input data of red, green, and blue can be converted into brightness, hue, and vividness. Features neural network color data conversion device.